{"aliases": [], "authors": ["Zhizheng Zhang", "Xiaoyi Zhang", "Wenxuan Xie", "Yan Lu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "631a39d2929c68bb93a7f95af22c17d2ab208668fd97d9082fe0601a73d123dd", "submitted_at": "2023-06-02T02:42:58Z", "updated_at": "2023-12-04T13:36:59Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2306.01242v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:42:42Z", "role": "primary", "sha256": "d1c1bc3fee56d124427abd2aabbcf876418633a8adf7a03f4ef0317cba6a42ae", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2008.05132%2C2010.04295%2C2101.04893%2C2112.05692%2C2306.01242%2C2311.03079%2C2401.05459%2C2402.03746%2C2402.15506%2C2403.20329%2C2404.06395%2C2404.10179%2C2405.10292%2C2406.09187%2C2406.14596%2C2408.03326%2C2409.12089%2C2501.10893%2C2504.09352&max_results=19"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:42:42Z", "role": "primary", "sha256": "d1c1bc3fee56d124427abd2aabbcf876418633a8adf7a03f4ef0317cba6a42ae", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2008.05132%2C2010.04295%2C2101.04893%2C2112.05692%2C2306.01242%2C2311.03079%2C2401.05459%2C2402.03746%2C2402.15506%2C2403.20329%2C2404.06395%2C2404.10179%2C2405.10292%2C2406.09187%2C2406.14596%2C2408.03326%2C2409.12089%2C2501.10893%2C2504.09352&max_results=19"}], "id": "arxiv:2306.01242", "identifiers": {"arxiv": "2306.01242"}, "kind": "paper", "last_verified_at": "2026-10-16T06:42:42Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2306.01242"}], "name": "Responsible Task Automation: Empowering Large Language Models as Responsible Task Automators", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced. UI task automation is stated in the abstract, but its platform and detailed evaluation protocol are not identified.", "published_at": "2023-06-02", "relations": [], "review_status": "curated", "scope": "adjacent", "summary": "Presents ResponsibleTA, a coordinator-executor framework that checks command feasibility and task completion while using local memory for privacy-related controls. The paper compares prompted and learned implementations and evaluates them on UI task automation.", "topics": ["safety-security", "evaluation", "automation"], "updated_at": "2023-12-04", "url": "https://arxiv.org/abs/2306.01242"}
{"aliases": [], "authors": ["Lingbo Mo", "Zeyi Liao", "Boyuan Zheng", "Yu Su", "Chaowei Xiao", "Huan Sun"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "b6d81a6b9de6fa32c8f5749d029d0819ddf3b310ae00a4872877ee72580e8879", "submitted_at": "2024-02-15T18:51:32Z", "updated_at": "2024-02-15T18:51:32Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2402.10196v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:08Z", "role": "primary", "sha256": "00d275cd5407cec847fa9d43fe73cfc4384c5570b68a5e058c97921799be0bec", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2302.01560%2C2303.16563%2C2305.15486%2C2305.16291%2C2305.17144%2C2306.00003%2C2306.00349%2C2306.17882%2C2310.01557%2C2310.08235%2C2310.08367%2C2311.05997%2C2311.17600%2C2312.02519%2C2312.11865%2C2402.01118%2C2402.10196%2C2402.11208%2C2403.10249%2C2403.12037%2C2404.06647%2C2404.12241%2C2404.13246%2C2407.15325%2C2409.00138%2C2410.14573%2C2410.18927%2C2410.19012%2C2411.09523%2C2411.10349%2C2411.13543%2C2412.01547%2C2501.05366%2C2501.09004%2C2502.07463%2C2502.14538%2C2503.02732%2C2503.08175%2C2503.11387%2C2504.10147%2C2504.11703%2C2504.14064%2C2504.15810%2C2504.16208%2C2504.20635%2C2504.21776%2C2505.12981%2C2505.13076%2C2506.01616%2C2506.13205&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:08Z", "role": "primary", "sha256": "00d275cd5407cec847fa9d43fe73cfc4384c5570b68a5e058c97921799be0bec", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2302.01560%2C2303.16563%2C2305.15486%2C2305.16291%2C2305.17144%2C2306.00003%2C2306.00349%2C2306.17882%2C2310.01557%2C2310.08235%2C2310.08367%2C2311.05997%2C2311.17600%2C2312.02519%2C2312.11865%2C2402.01118%2C2402.10196%2C2402.11208%2C2403.10249%2C2403.12037%2C2404.06647%2C2404.12241%2C2404.13246%2C2407.15325%2C2409.00138%2C2410.14573%2C2410.18927%2C2410.19012%2C2411.09523%2C2411.10349%2C2411.13543%2C2412.01547%2C2501.05366%2C2501.09004%2C2502.07463%2C2502.14538%2C2503.02732%2C2503.08175%2C2503.11387%2C2504.10147%2C2504.11703%2C2504.14064%2C2504.15810%2C2504.16208%2C2504.20635%2C2504.21776%2C2505.12981%2C2505.13076%2C2506.01616%2C2506.13205&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T13:00:03.117157+00:00", "role": "primary", "sha256": "a1d4dc44a6572863df1d8cc68bff592c3a04a174a3dbc87ab5cdcd0d33be92b0", "supports": ["summary", "scope", "environments", "notes"], "url": "https://arxiv.org/html/2402.10196v1"}], "id": "arxiv:2402.10196", "identifiers": {"arxiv": "2402.10196"}, "kind": "survey", "last_verified_at": "2026-10-18T10:54:08Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2402.10196"}], "name": "A Trembling House of Cards? Mapping Adversarial Attacks against Language Agents", "notes": "Adjacent conceptual security work: the paper describes hypothetical scenarios and a future generalist agent, rather than an implemented attack benchmark. Scope was checked against the retained abstract and full-text Sections 2.4 and 3; no results were independently reproduced.", "published_at": "2024-02-15", "relations": [], "review_status": "curated", "scope": "adjacent", "summary": "Presents a conceptual perception–reasoning–action framework and 12 potential attack scenarios for language agents, including hypothetical web-navigation attacks.", "topics": ["safety-security", "survey", "robustness"], "updated_at": "2024-02-15", "url": "https://arxiv.org/abs/2402.10196"}
{"aliases": [], "authors": ["Fangzhou Wu", "Shutong Wu", "Yulong Cao", "Chaowei Xiao"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "2cf1458c8c4134b2195394a8acba4919f9b3a5a7c79b1bea017ac4558123f69c", "submitted_at": "2024-02-26T19:01:54Z", "updated_at": "2024-02-26T19:01:54Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2402.16965v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2402.16965", "identifiers": {"arxiv": "2402.16965"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:58Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2402.16965"}], "name": "WIPI: A New Web Threat for LLM-Driven Web Agents", "notes": "Reviewed against the retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2024-02-26", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies WIPI, a black-box indirect-instruction threat in which webpage content redirects LLM-driven web agents. Evaluation spans plugin-based agents, Web GPTs, and open-source web-agent implementations.", "topics": ["safety-security", "web-navigation", "evaluation"], "updated_at": "2024-02-26", "url": "https://arxiv.org/abs/2402.16965"}
{"aliases": [], "authors": ["Zhen Xiang", "Linzhi Zheng", "Yanjie Li", "Junyuan Hong", "Qinbin Li", "Han Xie", "Jiawei Zhang", "Zidi Xiong", "Chulin Xie", "Carl Yang", "Dawn Song", "Bo Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "50a8ea492d3f1c2ff2bd00ebfb18e19077bacc96b44c751d985bfe0f7bef0247", "submitted_at": "2024-06-13T14:49:26Z", "updated_at": "2025-05-29T03:09:05Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2406.09187v3"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:42:42Z", "role": "primary", "sha256": "d1c1bc3fee56d124427abd2aabbcf876418633a8adf7a03f4ef0317cba6a42ae", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2008.05132%2C2010.04295%2C2101.04893%2C2112.05692%2C2306.01242%2C2311.03079%2C2401.05459%2C2402.03746%2C2402.15506%2C2403.20329%2C2404.06395%2C2404.10179%2C2405.10292%2C2406.09187%2C2406.14596%2C2408.03326%2C2409.12089%2C2501.10893%2C2504.09352&max_results=19"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:42:42Z", "role": "primary", "sha256": "d1c1bc3fee56d124427abd2aabbcf876418633a8adf7a03f4ef0317cba6a42ae", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2008.05132%2C2010.04295%2C2101.04893%2C2112.05692%2C2306.01242%2C2311.03079%2C2401.05459%2C2402.03746%2C2402.15506%2C2403.20329%2C2404.06395%2C2404.10179%2C2405.10292%2C2406.09187%2C2406.14596%2C2408.03326%2C2409.12089%2C2501.10893%2C2504.09352&max_results=19"}], "id": "arxiv:2406.09187", "identifiers": {"arxiv": "2406.09187"}, "kind": "paper", "last_verified_at": "2026-10-16T06:42:42Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2406.09187"}], "name": "GuardAgent: Safeguard LLM Agents by a Guard Agent via Knowledge-Enabled Reasoning", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced. The web-agent evidence is Mind2Web-SC; the healthcare benchmark does not independently establish GUI control.", "published_at": "2024-06-13", "relations": [], "review_status": "curated", "scope": "adjacent", "summary": "Introduces GuardAgent, which translates safety requests into executable guardrail code using plans and retrieved demonstrations. Mind2Web-SC supplies a web-agent safety-policy evaluation alongside a separate healthcare access-control benchmark.", "topics": ["safety-security", "evaluation", "tool-integration"], "updated_at": "2025-05-29", "url": "https://arxiv.org/abs/2406.09187"}
{"aliases": [], "authors": ["Chen Henry Wu", "Rishi Shah", "Jing Yu Koh", "Ruslan Salakhutdinov", "Daniel Fried", "Aditi Raghunathan"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "7cc1c1ac93f8e328d33accf7cdaa1d65487c4f7abec15234b9e596a501947550", "submitted_at": "2024-06-18T17:32:48Z", "updated_at": "2025-02-04T20:02:17Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2406.12814v3"}, "publication": {"authors": ["Chen Wu", "Rishi Shah", "Jing Yu Koh", "Russ Salakhutdinov", "Daniel Fried", "Aditi Raghunathan"], "date_basis": "Explicit publisher citation_publication_date: 2025-05-01. Preserve supplied precision; do not substitute conference event dates.", "date_parts": [2025, 5, 1], "doi": null, "event_dates": null, "evidence_urls": ["https://proceedings.iclr.cc/paper_files/paper/2025/hash/460a1d8eac34125dad453b28d6d64446-Abstract-Conference.html"], "identity_basis": "Same work: normalized title, ordered author correspondence and the specific contribution in the complete publisher abstract were reviewed against the retained arXiv identity. The normalized publisher abstract matches the retained arXiv abstract SHA-256.", "issue": null, "notes": "Publisher proceedings edition; retained separately from the arXiv snapshot. Publisher spelling, initials, name annotations or author lists differ from the retained preprint.", "pages": "28362-28383", "pdf_url": "https://proceedings.iclr.cc/paper_files/paper/2025/file/460a1d8eac34125dad453b28d6d64446-Paper-Conference.pdf", "publisher": null, "series": null, "title": "Dissecting Adversarial Robustness of Multimodal LM Agents", "type": "conference-paper", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/460a1d8eac34125dad453b28d6d64446-Abstract-Conference.html", "venue": "International Conference on Learning Representations", "volume": "2025"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:31:32Z", "role": "primary", "sha256": "b9a83994d8ac36ccdb606a5219511f8bb35ad3948ade4f9f3e28d386221aedfb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2406.12814%2C2406.13719%2C2406.14056%2C2406.14250%2C2406.14314%2C2406.18082%2C2406.19263%2C2407.00993%2C2407.01476%2C2407.01489%2C2407.01511%2C2407.01558%2C2407.03037%2C2407.03913%2C2407.04346%2C2407.05291%2C2407.07061%2C2407.08850%2C2407.09018%2C2407.09295%2C2407.10956%2C2407.13032%2C2407.13063%2C2407.15673%2C2407.15711%2C2407.16741%2C2407.17490%2C2407.18901%2C2407.19056%2C2407.20183%2C2408.00203%2C2408.02544%2C2408.06327%2C2408.07199%2C2408.11824%2C2408.15978%2C2409.02453%2C2409.04566%2C2409.07429%2C2409.08264%2C2409.09354%2C2409.10741%2C2409.11295%2C2409.13711%2C2409.14337%2C2409.14818%2C2409.15441%2C2409.15637%2C2409.17140%2C2409.20566&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T04:10:01Z", "role": "primary", "sha256": "fa23fda1b80dac019741737fc6f611fccf9234a4d73ffbb8e24e54d8caf4dac3", "supports": ["details.publication.authors", "details.publication.date_basis", "details.publication.date_parts", "details.publication.identity_basis", "details.publication.notes", "details.publication.pages", "details.publication.pdf_url", "details.publication.title", "details.publication.type", "details.publication.url", "details.publication.venue", "details.publication.volume"], "url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/460a1d8eac34125dad453b28d6d64446-Abstract-Conference.html"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:31:32Z", "role": "primary", "sha256": "b9a83994d8ac36ccdb606a5219511f8bb35ad3948ade4f9f3e28d386221aedfb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2406.12814%2C2406.13719%2C2406.14056%2C2406.14250%2C2406.14314%2C2406.18082%2C2406.19263%2C2407.00993%2C2407.01476%2C2407.01489%2C2407.01511%2C2407.01558%2C2407.03037%2C2407.03913%2C2407.04346%2C2407.05291%2C2407.07061%2C2407.08850%2C2407.09018%2C2407.09295%2C2407.10956%2C2407.13032%2C2407.13063%2C2407.15673%2C2407.15711%2C2407.16741%2C2407.17490%2C2407.18901%2C2407.19056%2C2407.20183%2C2408.00203%2C2408.02544%2C2408.06327%2C2408.07199%2C2408.11824%2C2408.15978%2C2409.02453%2C2409.04566%2C2409.07429%2C2409.08264%2C2409.09354%2C2409.10741%2C2409.11295%2C2409.13711%2C2409.14337%2C2409.14818%2C2409.15441%2C2409.15637%2C2409.17140%2C2409.20566&max_results=50"}], "id": "arxiv:2406.12814", "identifiers": {"arxiv": "2406.12814"}, "kind": "paper", "last_verified_at": "2026-10-16T06:31:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2406.12814"}, {"rel": "publication", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/460a1d8eac34125dad453b28d6d64446-Abstract-Conference.html"}, {"rel": "publication_pdf", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/file/460a1d8eac34125dad453b28d6d64446-Paper-Conference.pdf"}], "name": "Dissecting Adversarial Robustness of Multimodal LM Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2024-06-18", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces adversarial tasks on VisualWebArena and an evaluation framework that tracks how manipulated information passes between agent components. The analysis examines vulnerabilities introduced by reflection and tree-search evaluators as well as direct action generation.", "topics": ["safety-security", "robustness", "evaluation"], "updated_at": "2025-02-04", "url": "https://arxiv.org/abs/2406.12814"}
{"aliases": [], "authors": ["Yulong Yang", "Xinshan Yang", "Shuaidong Li", "Chenhao Lin", "Zhengyu Zhao", "Chao Shen", "Tianwei Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "88576325385d3c361f0666311eae8858ea0d4a0098d73221f789550bebf1e5db", "submitted_at": "2024-07-12T14:30:05Z", "updated_at": "2025-03-16T07:13:53Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2407.09295v3"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2407.09295", "identifiers": {"arxiv": "2407.09295"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:58Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2407.09295"}], "name": "Systematic Categorization, Construction and Evaluation of New Attacks against Multi-modal Mobile GUI Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2024-07-12", "relations": [], "review_status": "curated", "scope": "core", "summary": "Develops a threat taxonomy and evaluation framework for attacks against multimodal mobile GUI agents, with case studies and dataset experiments.", "topics": ["safety-security", "evaluation"], "updated_at": "2025-03-16", "url": "https://arxiv.org/abs/2407.09295"}
{"aliases": [], "authors": ["Xinbei Ma", "Yiting Wang", "Yao Yao", "Tongxin Yuan", "Aston Zhang", "Zhuosheng Zhang", "Hai Zhao"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "d66c0c2102dd2734d5334faeb12b4971d50a2d2e50a6671ede2fa78a84a0c03f", "submitted_at": "2024-08-05T15:16:22Z", "updated_at": "2025-09-05T09:21:10Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2408.02544v3"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:31:32Z", "role": "primary", "sha256": "b9a83994d8ac36ccdb606a5219511f8bb35ad3948ade4f9f3e28d386221aedfb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2406.12814%2C2406.13719%2C2406.14056%2C2406.14250%2C2406.14314%2C2406.18082%2C2406.19263%2C2407.00993%2C2407.01476%2C2407.01489%2C2407.01511%2C2407.01558%2C2407.03037%2C2407.03913%2C2407.04346%2C2407.05291%2C2407.07061%2C2407.08850%2C2407.09018%2C2407.09295%2C2407.10956%2C2407.13032%2C2407.13063%2C2407.15673%2C2407.15711%2C2407.16741%2C2407.17490%2C2407.18901%2C2407.19056%2C2407.20183%2C2408.00203%2C2408.02544%2C2408.06327%2C2408.07199%2C2408.11824%2C2408.15978%2C2409.02453%2C2409.04566%2C2409.07429%2C2409.08264%2C2409.09354%2C2409.10741%2C2409.11295%2C2409.13711%2C2409.14337%2C2409.14818%2C2409.15441%2C2409.15637%2C2409.17140%2C2409.20566&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:31:32Z", "role": "primary", "sha256": "b9a83994d8ac36ccdb606a5219511f8bb35ad3948ade4f9f3e28d386221aedfb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2406.12814%2C2406.13719%2C2406.14056%2C2406.14250%2C2406.14314%2C2406.18082%2C2406.19263%2C2407.00993%2C2407.01476%2C2407.01489%2C2407.01511%2C2407.01558%2C2407.03037%2C2407.03913%2C2407.04346%2C2407.05291%2C2407.07061%2C2407.08850%2C2407.09018%2C2407.09295%2C2407.10956%2C2407.13032%2C2407.13063%2C2407.15673%2C2407.15711%2C2407.16741%2C2407.17490%2C2407.18901%2C2407.19056%2C2407.20183%2C2408.00203%2C2408.02544%2C2408.06327%2C2408.07199%2C2408.11824%2C2408.15978%2C2409.02453%2C2409.04566%2C2409.07429%2C2409.08264%2C2409.09354%2C2409.10741%2C2409.11295%2C2409.13711%2C2409.14337%2C2409.14818%2C2409.15441%2C2409.15637%2C2409.17140%2C2409.20566&max_results=50"}], "id": "arxiv:2408.02544", "identifiers": {"arxiv": "2408.02544"}, "kind": "paper", "last_verified_at": "2026-10-16T06:31:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2408.02544"}], "name": "Caution for the Environment: Multimodal LLM Agents are Susceptible to Environmental Distractions", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced. The abstract describes a simulated dataset; real-browser deployment is not asserted.", "published_at": "2024-08-05", "relations": [], "review_status": "curated", "scope": "core", "summary": "Measures how irrelevant GUI content affects multimodal agents using a simulated dataset and three perception patterns. The study also examines adversarial environment injection and possible ways to improve instruction faithfulness.", "topics": ["safety-security", "robustness", "evaluation"], "updated_at": "2025-09-05", "url": "https://arxiv.org/abs/2408.02544"}
{"aliases": [], "authors": ["Zeyi Liao", "Lingbo Mo", "Chejian Xu", "Mintong Kang", "Jiawei Zhang", "Chaowei Xiao", "Yuan Tian", "Bo Li", "Huan Sun"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "897cdbb6aa52478b260ecba830e0a13527c2614454f7c766c9f5c8c9bdbfd4e3", "submitted_at": "2024-09-17T15:49:44Z", "updated_at": "2025-03-12T20:54:00Z", "version": 5, "versioned_url": "https://arxiv.org/abs/2409.11295v5"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2409.11295", "identifiers": {"arxiv": "2409.11295"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:58Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2409.11295"}], "name": "EIA: Environmental Injection Attack on Generalist Web Agents for Privacy Leakage", "notes": "Reviewed against the retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2024-09-17", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates privacy-directed webpage instruction injection against generalist web agents using PII-related interaction steps from Mind2Web, including attempts to expose specific user information or entire requests.", "topics": ["safety-security", "web-navigation", "evaluation"], "updated_at": "2025-03-12", "url": "https://arxiv.org/abs/2409.11295"}
{"aliases": [], "authors": ["Ido Levy", "Ben Wiesel", "Sami Marreed", "Alon Oved", "Avi Yaeli", "Nir Mashkif", "Segev Shlomov"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "ba2b599c2b3f09974ee41d329545b69773a3b408b6db0124c5b1073d5fe8f1cf", "submitted_at": "2024-10-09T09:13:38Z", "updated_at": "2026-06-04T09:59:28Z", "version": 7, "versioned_url": "https://arxiv.org/abs/2410.06703v7"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2410.06703", "identifiers": {"arxiv": "2410.06703"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:58Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2410.06703"}], "name": "ST-WebAgentBench: A Benchmark for Evaluating Safety and Trustworthiness in Web Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2024-10-09", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates web-task completion under explicit enterprise policies, measuring policy-compliant completion and safety violations separately.", "topics": ["safety-security", "evaluation"], "updated_at": "2026-06-04", "url": "https://arxiv.org/abs/2410.06703"}
{"aliases": [], "authors": ["Zhuohao Jerry Zhang", "Eldon Schoop", "Jeffrey Nichols", "Anuj Mahajan", "Amanda Swearngin"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "ff649a0792864bd40b7bf2f825f6a8a815eb60cc795ed4dc0c6b55a942477758", "submitted_at": "2024-10-11T17:24:00Z", "updated_at": "2025-03-22T18:01:37Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2410.09006v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:31:35Z", "role": "primary", "sha256": "b3e67984f59a1b05ccd946eb54e2d37b0bb78ab0ca796fbb6d544793dafea3e0", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2410.00467%2C2410.00689%2C2410.02810%2C2410.02907%2C2410.05243%2C2410.06703%2C2410.08164%2C2410.09006%2C2410.09024%2C2410.11843%2C2410.11871%2C2410.11872%2C2410.13232%2C2410.13757%2C2410.13824%2C2410.13825%2C2410.13886%2C2410.14803%2C2410.15164%2C2410.16464%2C2410.17236%2C2410.17401%2C2410.17520%2C2410.17883%2C2410.18603%2C2410.18963%2C2410.18967%2C2410.19054%2C2410.19100%2C2410.19461%2C2410.19609%2C2410.20745%2C2410.22552%2C2410.23218%2C2410.24024%2C2411.00820%2C2411.02337%2C2411.02391%2C2411.04468%2C2411.04890%2C2411.06559%2C2411.07618%2C2411.10323%2C2411.13451%2C2411.13591%2C2411.15004%2C2411.17465%2C2411.18279%2C2412.00402%2C2412.01268&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:31:35Z", "role": "primary", "sha256": "b3e67984f59a1b05ccd946eb54e2d37b0bb78ab0ca796fbb6d544793dafea3e0", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2410.00467%2C2410.00689%2C2410.02810%2C2410.02907%2C2410.05243%2C2410.06703%2C2410.08164%2C2410.09006%2C2410.09024%2C2410.11843%2C2410.11871%2C2410.11872%2C2410.13232%2C2410.13757%2C2410.13824%2C2410.13825%2C2410.13886%2C2410.14803%2C2410.15164%2C2410.16464%2C2410.17236%2C2410.17401%2C2410.17520%2C2410.17883%2C2410.18603%2C2410.18963%2C2410.18967%2C2410.19054%2C2410.19100%2C2410.19461%2C2410.19609%2C2410.20745%2C2410.22552%2C2410.23218%2C2410.24024%2C2411.00820%2C2411.02337%2C2411.02391%2C2411.04468%2C2411.04890%2C2411.06559%2C2411.07618%2C2411.10323%2C2411.13451%2C2411.13591%2C2411.15004%2C2411.17465%2C2411.18279%2C2412.00402%2C2412.01268&max_results=50"}], "id": "arxiv:2410.09006", "identifiers": {"arxiv": "2410.09006"}, "kind": "paper", "last_verified_at": "2026-10-16T06:31:35Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2410.09006"}], "name": "From Interaction to Impact: Towards Safer AI Agents Through Understanding and Evaluating Mobile UI Operation Impacts", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2024-10-11", "relations": [], "review_status": "curated", "scope": "core", "summary": "Develops a taxonomy of mobile-action consequences and annotated interaction data to evaluate whether models recognize impactful or risky UI operations.", "topics": ["safety-security", "mobile-navigation", "evaluation"], "updated_at": "2025-03-22", "url": "https://arxiv.org/abs/2410.09006"}
{"aliases": [], "authors": ["Priyanshu Kumar", "Elaine Lau", "Saranya Vijayakumar", "Tu Trinh", "Scale Red Team", "Elaine Chang", "Vaughn Robinson", "Sean Hendryx", "Shuyan Zhou", "Matt Fredrikson", "Summer Yue", "Zifan Wang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "4922d9362b23bc0e41d47dacf590c6e9bd6e88cfa6418cf3517814131295270c", "submitted_at": "2024-10-11T06:54:12Z", "updated_at": "2024-10-21T18:06:58Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2410.13886v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2410.13886", "identifiers": {"arxiv": "2410.13886"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:58Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2410.13886"}], "name": "Refusal-Trained LLMs Are Easily Jailbroken As Browser Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2024-10-11", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces BrowserART and studies whether chat-model refusal behavior transfers to browser agents carrying out potentially harmful instructions.", "topics": ["safety-security", "evaluation"], "updated_at": "2024-10-21", "url": "https://arxiv.org/abs/2410.13886"}
{"aliases": [], "authors": ["Chejian Xu", "Mintong Kang", "Jiawei Zhang", "Zeyi Liao", "Lingbo Mo", "Mengqi Yuan", "Huan Sun", "Bo Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "226a4acbbaab42c8cbc44a897651d8c745119426ea1223f1b849558af32fab02", "submitted_at": "2024-10-22T20:18:26Z", "updated_at": "2025-05-31T18:34:01Z", "version": 4, "versioned_url": "https://arxiv.org/abs/2410.17401v4"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2410.17401", "identifiers": {"arxiv": "2410.17401"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:58Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2410.17401"}], "name": "AdvAgent: Controllable Blackbox Red-teaming on Web Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2024-10-22", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces a black-box web-agent red-teaming framework that learns adversarial prompts from agent feedback to study security weaknesses.", "topics": ["safety-security", "evaluation"], "updated_at": "2025-05-31", "url": "https://arxiv.org/abs/2410.17401"}
{"aliases": [], "authors": ["Juyong Lee", "Dongyoon Hahm", "June Suk Choi", "W. Bradley Knox", "Kimin Lee"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "362cbf7ea267ddc4a64505aa95c4f16a9f3631f95319ce10eef9981e2c85937b", "submitted_at": "2024-10-23T02:51:43Z", "updated_at": "2026-01-27T01:51:55Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2410.17520v3"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:31:35Z", "role": "primary", "sha256": "b3e67984f59a1b05ccd946eb54e2d37b0bb78ab0ca796fbb6d544793dafea3e0", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2410.00467%2C2410.00689%2C2410.02810%2C2410.02907%2C2410.05243%2C2410.06703%2C2410.08164%2C2410.09006%2C2410.09024%2C2410.11843%2C2410.11871%2C2410.11872%2C2410.13232%2C2410.13757%2C2410.13824%2C2410.13825%2C2410.13886%2C2410.14803%2C2410.15164%2C2410.16464%2C2410.17236%2C2410.17401%2C2410.17520%2C2410.17883%2C2410.18603%2C2410.18963%2C2410.18967%2C2410.19054%2C2410.19100%2C2410.19461%2C2410.19609%2C2410.20745%2C2410.22552%2C2410.23218%2C2410.24024%2C2411.00820%2C2411.02337%2C2411.02391%2C2411.04468%2C2411.04890%2C2411.06559%2C2411.07618%2C2411.10323%2C2411.13451%2C2411.13591%2C2411.15004%2C2411.17465%2C2411.18279%2C2412.00402%2C2412.01268&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:31:35Z", "role": "primary", "sha256": "b3e67984f59a1b05ccd946eb54e2d37b0bb78ab0ca796fbb6d544793dafea3e0", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2410.00467%2C2410.00689%2C2410.02810%2C2410.02907%2C2410.05243%2C2410.06703%2C2410.08164%2C2410.09006%2C2410.09024%2C2410.11843%2C2410.11871%2C2410.11872%2C2410.13232%2C2410.13757%2C2410.13824%2C2410.13825%2C2410.13886%2C2410.14803%2C2410.15164%2C2410.16464%2C2410.17236%2C2410.17401%2C2410.17520%2C2410.17883%2C2410.18603%2C2410.18963%2C2410.18967%2C2410.19054%2C2410.19100%2C2410.19461%2C2410.19609%2C2410.20745%2C2410.22552%2C2410.23218%2C2410.24024%2C2411.00820%2C2411.02337%2C2411.02391%2C2411.04468%2C2411.04890%2C2411.06559%2C2411.07618%2C2411.10323%2C2411.13451%2C2411.13591%2C2411.15004%2C2411.17465%2C2411.18279%2C2412.00402%2C2412.01268&max_results=50"}], "id": "arxiv:2410.17520", "identifiers": {"arxiv": "2410.17520"}, "kind": "paper", "last_verified_at": "2026-10-16T06:31:35Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2410.17520"}], "name": "MobileSafetyBench: Evaluating Safety of Autonomous Agents in Mobile Device Control", "notes": "Summary and topic tags reviewed against the primary abstract. Reported experiments, task counts, and artifact availability are not independently reproduced.", "published_at": "2024-10-23", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates mobile agents on misuse, side effects, and indirect prompt-injection risks in an Android emulator environment.", "topics": ["safety-security", "evaluation"], "updated_at": "2026-01-27", "url": "https://arxiv.org/abs/2410.17520"}
{"aliases": [], "authors": ["Haoyi Qiu", "Alexander R. Fabbri", "Divyansh Agarwal", "Kung-Hsiang Huang", "Sarah Tan", "Nanyun Peng", "Chien-Sheng Wu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "26fac89c78ab03fcbaf25bbfe93ebce090dec9adbf5c6dd5440641f154cffac9", "submitted_at": "2024-10-30T17:35:44Z", "updated_at": "2025-03-08T23:37:49Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2410.23252v3"}, "publication": {"authors": ["Haoyi Qiu", "Alexander Richard Fabbri", "Divyansh Agarwal", "Kung-Hsiang Huang", "Sarah Tan", "Nanyun Peng", "Chien-Sheng Wu"], "date_basis": "Explicit publisher citation_publication_date: 2025/4. Preserve supplied precision; do not substitute conference event dates.", "date_parts": [2025, 4], "doi": "10.18653/v1/2025.findings-naacl.222", "event_dates": null, "evidence_urls": ["https://aclanthology.org/2025.findings-naacl.222/"], "identity_basis": "Same work: normalized title, ordered author correspondence and the specific contribution in the complete publisher abstract were reviewed against the retained arXiv identity.", "issue": null, "notes": "Publisher proceedings edition; retained separately from the arXiv snapshot. Publisher spelling, initials, name annotations or author lists differ from the retained preprint.", "pages": "3978-4005", "pdf_url": "https://aclanthology.org/2025.findings-naacl.222.pdf", "publisher": "Association for Computational Linguistics", "series": null, "title": "Evaluating Cultural and Social Awareness of LLM Web Agents", "type": "conference-paper", "url": "https://aclanthology.org/2025.findings-naacl.222/", "venue": "Findings of the Association for Computational Linguistics: NAACL 2025", "volume": null}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T04:09:55Z", "role": "primary", "sha256": "6563b807356c30b69a1c5bb9328358c139d171777cb893079ddc123cb82a282e", "supports": ["details.publication.authors", "details.publication.date_basis", "details.publication.date_parts", "details.publication.doi", "details.publication.identity_basis", "details.publication.notes", "details.publication.pages", "details.publication.pdf_url", "details.publication.publisher", "details.publication.title", "details.publication.type", "details.publication.url", "details.publication.venue"], "url": "https://aclanthology.org/2025.findings-naacl.222/"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2410.23252", "identifiers": {"arxiv": "2410.23252"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:58Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2410.23252"}, {"rel": "publication", "url": "https://aclanthology.org/2025.findings-naacl.222/"}, {"rel": "publication_pdf", "url": "https://aclanthology.org/2025.findings-naacl.222.pdf"}], "name": "Evaluating Cultural and Social Awareness of LLM Web Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2024-10-30", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces CASA to assess web agents' responses to cultural and social norms in shopping and forum tasks, including misleading observations.", "topics": ["safety-security", "evaluation"], "updated_at": "2025-03-08", "url": "https://arxiv.org/abs/2410.23252"}
{"aliases": [], "authors": ["Yanzhe Zhang", "Tao Yu", "Diyi Yang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "085c64595365cc6439c8c5fa8a479259a3d59b7e9740e0f9d43e48d00e1439c0", "submitted_at": "2024-11-04T18:56:42Z", "updated_at": "2025-05-24T16:15:46Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2411.02391v2"}, "publication": {"authors": ["Yanzhe Zhang", "Tao Yu", "Diyi Yang"], "date_basis": "Explicit publisher citation_publication_date: 2025/7. Preserve supplied precision; do not substitute conference event dates.", "date_parts": [2025, 7], "doi": "10.18653/v1/2025.acl-long.411", "event_dates": null, "evidence_urls": ["https://aclanthology.org/2025.acl-long.411/"], "identity_basis": "Same work: normalized title, ordered author correspondence and the specific contribution in the complete publisher abstract were reviewed against the retained arXiv identity.", "issue": null, "notes": "Publisher proceedings edition; retained separately from the arXiv snapshot.", "pages": "8387-8401", "pdf_url": "https://aclanthology.org/2025.acl-long.411.pdf", "publisher": "Association for Computational Linguistics", "series": null, "title": "Attacking Vision-Language Computer Agents via Pop-ups", "type": "conference-paper", "url": "https://aclanthology.org/2025.acl-long.411/", "venue": "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)", "volume": null}}, "environments": ["desktop", "web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:31:35Z", "role": "primary", "sha256": "b3e67984f59a1b05ccd946eb54e2d37b0bb78ab0ca796fbb6d544793dafea3e0", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2410.00467%2C2410.00689%2C2410.02810%2C2410.02907%2C2410.05243%2C2410.06703%2C2410.08164%2C2410.09006%2C2410.09024%2C2410.11843%2C2410.11871%2C2410.11872%2C2410.13232%2C2410.13757%2C2410.13824%2C2410.13825%2C2410.13886%2C2410.14803%2C2410.15164%2C2410.16464%2C2410.17236%2C2410.17401%2C2410.17520%2C2410.17883%2C2410.18603%2C2410.18963%2C2410.18967%2C2410.19054%2C2410.19100%2C2410.19461%2C2410.19609%2C2410.20745%2C2410.22552%2C2410.23218%2C2410.24024%2C2411.00820%2C2411.02337%2C2411.02391%2C2411.04468%2C2411.04890%2C2411.06559%2C2411.07618%2C2411.10323%2C2411.13451%2C2411.13591%2C2411.15004%2C2411.17465%2C2411.18279%2C2412.00402%2C2412.01268&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T04:09:56Z", "role": "primary", "sha256": "a28a603d74e08954a05265f54c8765bd874a6cb2fc0c5a68a35e15e68692b0a1", "supports": ["details.publication.authors", "details.publication.date_basis", "details.publication.date_parts", "details.publication.doi", "details.publication.identity_basis", "details.publication.notes", "details.publication.pages", "details.publication.pdf_url", "details.publication.publisher", "details.publication.title", "details.publication.type", "details.publication.url", "details.publication.venue"], "url": "https://aclanthology.org/2025.acl-long.411/"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:31:35Z", "role": "primary", "sha256": "b3e67984f59a1b05ccd946eb54e2d37b0bb78ab0ca796fbb6d544793dafea3e0", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2410.00467%2C2410.00689%2C2410.02810%2C2410.02907%2C2410.05243%2C2410.06703%2C2410.08164%2C2410.09006%2C2410.09024%2C2410.11843%2C2410.11871%2C2410.11872%2C2410.13232%2C2410.13757%2C2410.13824%2C2410.13825%2C2410.13886%2C2410.14803%2C2410.15164%2C2410.16464%2C2410.17236%2C2410.17401%2C2410.17520%2C2410.17883%2C2410.18603%2C2410.18963%2C2410.18967%2C2410.19054%2C2410.19100%2C2410.19461%2C2410.19609%2C2410.20745%2C2410.22552%2C2410.23218%2C2410.24024%2C2411.00820%2C2411.02337%2C2411.02391%2C2411.04468%2C2411.04890%2C2411.06559%2C2411.07618%2C2411.10323%2C2411.13451%2C2411.13591%2C2411.15004%2C2411.17465%2C2411.18279%2C2412.00402%2C2412.01268&max_results=50"}], "id": "arxiv:2411.02391", "identifiers": {"arxiv": "2411.02391"}, "kind": "paper", "last_verified_at": "2026-10-16T06:31:35Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2411.02391"}, {"rel": "publication", "url": "https://aclanthology.org/2025.acl-long.411/"}, {"rel": "publication_pdf", "url": "https://aclanthology.org/2025.acl-long.411.pdf"}], "name": "Attacking Vision-Language Computer Agents via Pop-ups", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2024-11-04", "relations": [], "review_status": "curated", "scope": "core", "summary": "Tests whether adversarial visual pop-ups divert computer agents from intended tasks in desktop and web benchmark environments.", "topics": ["safety-security", "evaluation"], "updated_at": "2025-05-24", "url": "https://arxiv.org/abs/2411.02391"}
{"aliases": [], "authors": ["Yurun Chen", "Xavier Hu", "Keting Yin", "Juncheng Li", "Shengyu Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "5af8d3a8b4db150bab0630f6040a0f3bced612f68b11d31985053f1f25c0f6f6", "submitted_at": "2025-02-18T17:01:28Z", "updated_at": "2025-08-06T07:04:43Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2502.13053v3"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:31:38Z", "role": "primary", "sha256": "e96b86015c763c780e0430a8d0867b96cae947f50e888b3441c85abcaa42de78", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2412.04454%2C2412.04626%2C2412.05467%2C2412.09362%2C2412.09605%2C2412.10342%2C2412.10742%2C2412.10840%2C2412.13103%2C2412.13194%2C2412.13501%2C2412.14161%2C2412.14470%2C2412.16256%2C2412.17589%2C2412.18116%2C2412.18426%2C2412.19723%2C2501.01149%2C2501.04575%2C2501.07572%2C2501.11733%2C2501.12326%2C2501.13896%2C2501.16150%2C2502.01600%2C2502.01977%2C2502.02982%2C2502.06395%2C2502.07949%2C2502.08047%2C2502.08226%2C2502.11357%2C2502.13053%2C2502.13923%2C2502.14282%2C2502.18906%2C2503.00401%2C2503.02268%2C2503.03196%2C2503.04730%2C2503.05143%2C2503.06470%2C2503.09263%2C2503.11069%2C2503.11170%2C2503.12687%2C2503.15661%2C2503.15937%2C2503.16465&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:31:38Z", "role": "primary", "sha256": "e96b86015c763c780e0430a8d0867b96cae947f50e888b3441c85abcaa42de78", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2412.04454%2C2412.04626%2C2412.05467%2C2412.09362%2C2412.09605%2C2412.10342%2C2412.10742%2C2412.10840%2C2412.13103%2C2412.13194%2C2412.13501%2C2412.14161%2C2412.14470%2C2412.16256%2C2412.17589%2C2412.18116%2C2412.18426%2C2412.19723%2C2501.01149%2C2501.04575%2C2501.07572%2C2501.11733%2C2501.12326%2C2501.13896%2C2501.16150%2C2502.01600%2C2502.01977%2C2502.02982%2C2502.06395%2C2502.07949%2C2502.08047%2C2502.08226%2C2502.11357%2C2502.13053%2C2502.13923%2C2502.14282%2C2502.18906%2C2503.00401%2C2503.02268%2C2503.03196%2C2503.04730%2C2503.05143%2C2503.06470%2C2503.09263%2C2503.11069%2C2503.11170%2C2503.12687%2C2503.15661%2C2503.15937%2C2503.16465&max_results=50"}], "id": "arxiv:2502.13053", "identifiers": {"arxiv": "2502.13053"}, "kind": "paper", "last_verified_at": "2026-10-16T06:31:38Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2502.13053"}], "name": "Evaluating the Robustness of Multimodal Agents Against Active Environmental Injection Attacks", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-02-18", "relations": [], "review_status": "curated", "scope": "core", "summary": "Examines active environmental injection against mobile agents, testing whether adversarial interface content redirects decisions during task execution.", "topics": ["safety-security", "robustness", "evaluation"], "updated_at": "2025-08-06", "url": "https://arxiv.org/abs/2502.13053"}
{"aliases": [], "authors": ["Jeffrey Yang Fan Chiang", "Seungjae Lee", "Jia-Bin Huang", "Furong Huang", "Yizheng Chen"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "27eacc2668411e2b7f79423940aa12293fa57aa81390c64587d3074c141dcb4f", "submitted_at": "2025-02-27T18:56:26Z", "updated_at": "2025-09-21T22:29:53Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2502.20383v3"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-17T07:57:44Z", "role": "primary", "sha256": "9ca057ee6538002c739e58a7b3f86fc51a434d085a33e5590f0720e4f27c6b24", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2501.06497%2C2501.06994%2C2501.07834%2C2501.07927%2C2501.11233%2C2501.15570%2C2501.16403%2C2501.16609%2C2501.17391%2C2502.00372%2C2502.00392%2C2502.00964%2C2502.00988%2C2502.01562%2C2502.01839%2C2502.03723%2C2502.05174%2C2502.05227%2C2502.08489%2C2502.11221%2C2502.12130%2C2502.14496%2C2502.16587%2C2502.16786%2C2502.16796%2C2502.18356%2C2502.18512%2C2502.20383%2C2502.20390%2C2502.20582%2C2503.06378%2C2503.08026%2C2503.10613%2C2503.10809%2C2503.11579%2C2503.12228%2C2503.12461%2C2503.13843%2C2503.14495%2C2503.15426%2C2503.16036%2C2503.16248%2C2503.16975%2C2503.17671%2C2503.18102%2C2503.21889%2C2503.21949%2C2503.23803%2C2504.01328%2C2504.01377&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T11:20:04Z", "role": "primary", "sha256": "0118540c95d316ed30aabf783c072d413677aaf89539316e08e1d81ddab2c5e6", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://arxiv.org/abs/2502.20383v3"}], "id": "arxiv:2502.20383", "identifiers": {"arxiv": "2502.20383"}, "kind": "paper", "last_verified_at": "2026-10-17T11:20:04Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2502.20383"}], "name": "Why Are Web AI Agents More Vulnerable Than Standalone LLMs? A Security Analysis", "notes": "Reviewed against the exact-version official arXiv abstract page; its complete title, ordered author list, visible version, and normalized abstract SHA match the retained snapshot. The original API response body was not locally available. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-02-27", "relations": [], "review_status": "curated", "scope": "core", "summary": "Analyzes web-agent vulnerabilities at the component level, examining goal placement in prompts, multistep action generation, and observations as sources of increased susceptibility.", "topics": ["safety-security", "evaluation", "web-navigation"], "updated_at": "2025-09-21", "url": "https://arxiv.org/abs/2502.20383"}
{"aliases": [], "authors": ["Ada Defne Tur", "Nicholas Meade", "Xing Han Lù", "Alejandra Zambrano", "Arkil Patel", "Esin Durmus", "Spandana Gella", "Karolina Stańczak", "Siva Reddy"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e4588022dd47624058db8a16d7f7763c591241298e94594b56bfeb4ab2e01737", "submitted_at": "2025-03-06T20:43:14Z", "updated_at": "2025-03-06T20:43:14Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2503.04957v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2503.04957", "identifiers": {"arxiv": "2503.04957"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2503.04957"}], "name": "SafeArena: Evaluating the Safety of Autonomous Web Agents", "notes": "Summary and topic tags reviewed against the primary abstract. Reported experiments, task counts, and artifact availability are not independently reproduced.", "published_at": "2025-03-06", "relations": [], "review_status": "curated", "scope": "core", "summary": "Pairs benign and harmful web tasks to assess deliberate misuse risks and categorize agent behavior by risk level.", "topics": ["safety-security", "evaluation"], "updated_at": "2025-03-06", "url": "https://arxiv.org/abs/2503.04957"}
{"aliases": [], "authors": ["Pei Yang", "Hai Ci", "Mike Zheng Shou"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "7d4423ee1e7047a6228f5b29684618fc4108af0694484a7ceddf6c3b53fbb06c", "submitted_at": "2025-03-12T10:38:15Z", "updated_at": "2025-03-12T10:38:15Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2503.09241v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T04:21:52Z", "role": "primary", "sha256": "c194d15b7e422aae48b4e05216a1f1d84471b6666718b9d3dd4c9d250c356e77", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2210.03347%2C2403.04984%2C2405.07090%2C2406.06282%2C2409.00088%2C2409.10811%2C2411.10943%2C2503.09241%2C2503.20279%2C2503.23350%2C2505.05440%2C2505.10593%2C2505.16422%2C2505.21279%2C2505.24878%2C2506.02153%2C2506.02839%2C2506.05079%2C2506.07976%2C2508.02046%2C2511.07885%2C2511.11248%2C2512.21720%2C2603.20351%2C2604.19299&max_results=25"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T04:21:52Z", "role": "primary", "sha256": "c194d15b7e422aae48b4e05216a1f1d84471b6666718b9d3dd4c9d250c356e77", "supports": ["name", "kind", "scope", "environments", "topics", "summary", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2210.03347%2C2403.04984%2C2405.07090%2C2406.06282%2C2409.00088%2C2409.10811%2C2411.10943%2C2503.09241%2C2503.20279%2C2503.23350%2C2505.05440%2C2505.10593%2C2505.16422%2C2505.21279%2C2505.24878%2C2506.02153%2C2506.02839%2C2506.05079%2C2506.07976%2C2508.02046%2C2511.07885%2C2511.11248%2C2512.21720%2C2603.20351%2C2604.19299&max_results=25"}], "id": "arxiv:2503.09241", "identifiers": {"arxiv": "2503.09241"}, "kind": "paper", "last_verified_at": "2026-10-18T04:21:52Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2503.09241"}], "name": "In-Context Defense in Computer Agents: An Empirical Study", "notes": "Reviewed against the complete official arXiv abstract. Experimental results and linked artifact availability have not been independently reproduced.", "published_at": "2025-03-12", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies in-context defenses for computer agents exposed to misleading pop-ups and other environmental instructions. It examines defensive reasoning before action planning and how a small set of examples changes resistance to these attacks.", "topics": ["safety-security", "evaluation"], "updated_at": "2025-03-12", "url": "https://arxiv.org/abs/2503.09241"}
{"aliases": [], "authors": ["Arman Zharmagambetov", "Chuan Guo", "Ivan Evtimov", "Maya Pavlova", "Ruslan Salakhutdinov", "Kamalika Chaudhuri"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "bbb0726446c9679781b8e5930da0d3f406989a4b47759ce3440c9ad56d5839d8", "submitted_at": "2025-03-12T19:30:31Z", "updated_at": "2025-10-01T23:55:58Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2503.09780v3"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2503.09780", "identifiers": {"arxiv": "2503.09780"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2503.09780"}], "name": "AgentDAM: Privacy Leakage Evaluation for Autonomous Web Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-03-12", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates whether autonomous web agents minimize use of sensitive personal information when completing tasks that expose unnecessary private data.", "topics": ["safety-security", "evaluation"], "updated_at": "2025-10-01", "url": "https://arxiv.org/abs/2503.09780"}
{"aliases": [], "authors": ["Piero A. Bonatti", "John Domingue", "Anna Lisa Gentile", "Andreas Harth", "Olaf Hartig", "Aidan Hogan", "Katja Hose", "Ernesto Jimenez-Ruiz", "Deborah L. McGuinness", "Chang Sun", "Ruben Verborgh", "Jesse Wright"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "4c778da33baa88916fbcb3e02a33f760dccea8b193d25087df9b56cd42c3c248", "submitted_at": "2025-01-31T12:26:27Z", "updated_at": "2025-01-31T12:26:27Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2503.15515v1"}}, "environments": ["web", "cross-platform"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:58Z", "role": "primary", "sha256": "c87e97eb397da04806d53145cf9c13be739fc657bb54dca9318c8c4ed2a36249", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=800&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2503.15515", "identifiers": {"arxiv": "2503.15515"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:58Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2503.15515"}], "name": "Towards Computer-Using Personal Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-01-31", "relations": [], "review_status": "curated", "scope": "core", "summary": "Proposes computer-using personal agents connected to user-controlled data repositories, discussing privacy, interoperability, and collaborative tasks.", "topics": ["safety-security", "memory", "human-in-the-loop"], "updated_at": "2025-01-31", "url": "https://arxiv.org/abs/2503.15515"}
{"aliases": [], "authors": ["Pengzhou Cheng", "Zheng Wu", "Zongru Wu", "Aston Zhang", "Zhuosheng Zhang", "Gongshen Liu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "aa6201528b13c6a1aae8dcd745f4f5d995e79982ef7c54a1962e2f9035233293", "submitted_at": "2025-02-26T12:31:16Z", "updated_at": "2025-07-14T15:56:44Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2503.16465v3"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2503.16465", "identifiers": {"arxiv": "2503.16465"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2503.16465"}], "name": "OS-Kairos: Adaptive Interaction for MLLM-Powered GUI Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-02-26", "relations": [], "review_status": "curated", "scope": "core", "summary": "Uses step-level confidence estimates to decide when a GUI agent should act autonomously or request human intervention in complex situations.", "topics": ["human-in-the-loop", "safety-security", "mobile-navigation"], "updated_at": "2025-07-14", "url": "https://arxiv.org/abs/2503.16465"}
{"aliases": [], "authors": ["Jungjae Lee", "Dongjae Lee", "Chihun Choi", "Youngmin Im", "Jaeyoung Wi", "Kihong Heo", "Sangeun Oh", "Sunjae Lee", "Insik Shin"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "10e7a9a4471555c6e4001cd10c1e4b375b503adfe6ea48d5808d1d7721dfc162", "submitted_at": "2025-03-24T09:46:05Z", "updated_at": "2025-09-11T04:15:45Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2503.18492v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2503.18492", "identifiers": {"arxiv": "2503.18492"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2503.18492"}], "name": "VeriSafe Agent: Safeguarding Mobile GUI Agent via Logic-based Action Verification", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-03-24", "relations": [], "review_status": "curated", "scope": "core", "summary": "Translates user instructions into formal specifications and applies runtime logic-based checks to proposed mobile GUI actions.", "topics": ["safety-security", "mobile-navigation"], "updated_at": "2025-09-11", "url": "https://arxiv.org/abs/2503.18492"}
{"aliases": [], "authors": ["Sejin Lee", "Jian Kim", "Haon Park", "Ashkan Yousefpour", "Sangyoon Yu", "Min Song"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "5e90cbc547703a2673cfb30bbe9079496ce4c076b51f39e0557f5db9048947c8", "submitted_at": "2025-03-26T07:08:15Z", "updated_at": "2025-06-08T05:39:56Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2503.20279v3"}}, "environments": ["desktop", "web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T04:21:52Z", "role": "primary", "sha256": "c194d15b7e422aae48b4e05216a1f1d84471b6666718b9d3dd4c9d250c356e77", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2210.03347%2C2403.04984%2C2405.07090%2C2406.06282%2C2409.00088%2C2409.10811%2C2411.10943%2C2503.09241%2C2503.20279%2C2503.23350%2C2505.05440%2C2505.10593%2C2505.16422%2C2505.21279%2C2505.24878%2C2506.02153%2C2506.02839%2C2506.05079%2C2506.07976%2C2508.02046%2C2511.07885%2C2511.11248%2C2512.21720%2C2603.20351%2C2604.19299&max_results=25"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T04:21:52Z", "role": "primary", "sha256": "c194d15b7e422aae48b4e05216a1f1d84471b6666718b9d3dd4c9d250c356e77", "supports": ["name", "kind", "scope", "environments", "topics", "summary", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2210.03347%2C2403.04984%2C2405.07090%2C2406.06282%2C2409.00088%2C2409.10811%2C2411.10943%2C2503.09241%2C2503.20279%2C2503.23350%2C2505.05440%2C2505.10593%2C2505.16422%2C2505.21279%2C2505.24878%2C2506.02153%2C2506.02839%2C2506.05079%2C2506.07976%2C2508.02046%2C2511.07885%2C2511.11248%2C2512.21720%2C2603.20351%2C2604.19299&max_results=25"}], "id": "arxiv:2503.20279", "identifiers": {"arxiv": "2503.20279"}, "kind": "paper", "last_verified_at": "2026-10-18T04:21:52Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2503.20279"}], "name": "sudo rm -rf agentic_security", "notes": "Reviewed against the complete official arXiv abstract. Experimental results and linked artifact availability have not been independently reproduced.", "published_at": "2025-03-26", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates security failures of computer-use agents under adversarial requests and screen-context manipulation. The study examines refusal behavior while agents perform tasks in desktop and web environments.", "topics": ["safety-security", "evaluation"], "updated_at": "2025-06-08", "url": "https://arxiv.org/abs/2503.20279"}
{"aliases": [], "authors": ["Zhaorun Chen", "Mintong Kang", "Bo Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "362fe117797181225641b71b23df95d82c2f90ebf2bbfa1bebea56427e502091", "submitted_at": "2025-03-26T17:58:40Z", "updated_at": "2025-11-27T22:01:17Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2503.22738v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T07:54:03Z", "role": "primary", "sha256": "13d5969845070cd2877dc51ae5300c2d623e5daef00c8a1552ffa46af3f721a5", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2401.05507%2C2406.07275%2C2409.16427%2C2410.00079%2C2503.07701%2C2503.22738%2C2505.15859%2C2505.17735%2C2505.23596%2C2507.05257%2C2507.06134%2C2507.21953%2C2507.23336%2C2508.06960%2C2508.08997%2C2509.02473%2C2509.23694%2C2510.09116%2C2510.14438%2C2511.18423%2C2512.12818%2C2512.13564%2C2601.10758%2C2601.12449%2C2601.21841%2C2603.01152%2C2603.19993%2C2603.20576%2C2604.04503%2C2604.17562%2C2604.23941%2C2605.16265%2C2606.06448%2C2606.06462%2C2606.29193%2C2607.01647%2C2607.13157&max_results=37"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T07:54:03Z", "role": "primary", "sha256": "13d5969845070cd2877dc51ae5300c2d623e5daef00c8a1552ffa46af3f721a5", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2401.05507%2C2406.07275%2C2409.16427%2C2410.00079%2C2503.07701%2C2503.22738%2C2505.15859%2C2505.17735%2C2505.23596%2C2507.05257%2C2507.06134%2C2507.21953%2C2507.23336%2C2508.06960%2C2508.08997%2C2509.02473%2C2509.23694%2C2510.09116%2C2510.14438%2C2511.18423%2C2512.12818%2C2512.13564%2C2601.10758%2C2601.12449%2C2601.21841%2C2603.01152%2C2603.19993%2C2603.20576%2C2604.04503%2C2604.17562%2C2604.23941%2C2605.16265%2C2606.06448%2C2606.06462%2C2606.29193%2C2607.01647%2C2607.13157&max_results=37"}], "id": "arxiv:2503.22738", "identifiers": {"arxiv": "2503.22738"}, "kind": "paper", "last_verified_at": "2026-10-16T07:54:03Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2503.22738"}], "name": "ShieldAgent: Shielding Agents via Verifiable Safety Policy Reasoning", "notes": "Editorial summary based on the primary paper abstract; experiments and released artifacts have not been independently reproduced.", "published_at": "2025-03-26", "relations": [], "review_status": "curated", "scope": "core", "summary": "Proposes a guardrail agent that reasons over explicit safety policies and checks action trajectories, with a web-agent safety benchmark.", "topics": ["safety-security"], "updated_at": "2025-11-27", "url": "https://arxiv.org/abs/2503.22738"}
{"aliases": [], "authors": ["Liangbo Ning", "Ziran Liang", "Zhuohang Jiang", "Haohao Qu", "Yujuan Ding", "Wenqi Fan", "Xiao-yong Wei", "Shanru Lin", "Hui Liu", "Philip S. Yu", "Qing Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "4b214bd8e5a34c0e8ca16a3be4fcf638a59a17ca524276dbef3c6b75f08398bb", "submitted_at": "2025-03-30T08:15:44Z", "updated_at": "2025-08-05T11:47:41Z", "version": 4, "versioned_url": "https://arxiv.org/abs/2503.23350v4"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T04:21:52Z", "role": "primary", "sha256": "c194d15b7e422aae48b4e05216a1f1d84471b6666718b9d3dd4c9d250c356e77", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2210.03347%2C2403.04984%2C2405.07090%2C2406.06282%2C2409.00088%2C2409.10811%2C2411.10943%2C2503.09241%2C2503.20279%2C2503.23350%2C2505.05440%2C2505.10593%2C2505.16422%2C2505.21279%2C2505.24878%2C2506.02153%2C2506.02839%2C2506.05079%2C2506.07976%2C2508.02046%2C2511.07885%2C2511.11248%2C2512.21720%2C2603.20351%2C2604.19299&max_results=25"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T04:21:52Z", "role": "primary", "sha256": "c194d15b7e422aae48b4e05216a1f1d84471b6666718b9d3dd4c9d250c356e77", "supports": ["name", "kind", "scope", "environments", "topics", "summary", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2210.03347%2C2403.04984%2C2405.07090%2C2406.06282%2C2409.00088%2C2409.10811%2C2411.10943%2C2503.09241%2C2503.20279%2C2503.23350%2C2505.05440%2C2505.10593%2C2505.16422%2C2505.21279%2C2505.24878%2C2506.02153%2C2506.02839%2C2506.05079%2C2506.07976%2C2508.02046%2C2511.07885%2C2511.11248%2C2512.21720%2C2603.20351%2C2604.19299&max_results=25"}], "id": "arxiv:2503.23350", "identifiers": {"arxiv": "2503.23350"}, "kind": "survey", "last_verified_at": "2026-10-18T04:21:52Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2503.23350"}], "name": "A Survey of WebAgents: Towards Next-Generation AI Agents for Web Automation with Large Foundation Models", "notes": "Reviewed against the complete official arXiv abstract. Experimental results and linked artifact availability have not been independently reproduced.", "published_at": "2025-03-30", "relations": [], "review_status": "curated", "scope": "core", "summary": "Surveys foundation-model web agents through their architectures, training methods and trustworthiness. It organizes research on completing browser tasks from user instructions and identifies directions for further development.", "topics": ["survey", "web-navigation", "safety-security"], "updated_at": "2025-08-05", "url": "https://arxiv.org/abs/2503.23350"}
{"aliases": [], "authors": ["Yucheng Shi", "Wenhao Yu", "Jingyuan Huang", "Wenlin Yao", "Wenhu Chen", "Ninghao Liu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "5a6811c6d0c3b30d4e6af2f7b0beef8aa8f1ad99ceb325182305e571fe5c398f", "submitted_at": "2025-03-30T13:26:00Z", "updated_at": "2026-02-24T05:24:23Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2503.23434v2"}}, "environments": ["cross-platform"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2503.23434", "identifiers": {"arxiv": "2503.23434"}, "kind": "survey", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2503.23434"}], "name": "Towards Trustworthy GUI Agents: A Survey", "notes": "", "published_at": "2025-03-30", "relations": [], "review_status": "curated", "scope": "core", "summary": "Frames GUI-agent trustworthiness as an execution gap and reviews failures, attacks, and defenses across perception, reasoning, and interaction.", "topics": ["survey", "safety-security"], "updated_at": "2026-02-24", "url": "https://arxiv.org/abs/2503.23434"}
{"aliases": [], "authors": ["Haoren Zhao", "Tianyi Chen", "Zhen Wang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "ededfa864150f1b7b4ba53b265b5cbaa60bcf42690a8e22f5c13f915b6cd0108", "submitted_at": "2025-04-07T03:58:45Z", "updated_at": "2025-04-07T03:58:45Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2504.04716v1"}}, "environments": ["web", "desktop", "cross-platform"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2504.04716", "identifiers": {"arxiv": "2504.04716"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2504.04716"}], "name": "On the Robustness of GUI Grounding Models Against Image Attacks", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-04-07", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates GUI grounding models under natural image noise, untargeted adversarial perturbations, and targeted image attacks. Experiments cover mobile, desktop, and web interfaces and analyze sensitivity to image resolution and adversarial changes.", "topics": ["grounding", "robustness", "safety-security", "evaluation"], "updated_at": "2025-04-07", "url": "https://arxiv.org/abs/2504.04716"}
{"aliases": [], "authors": ["Chaoran Chen", "Zhiping Zhang", "Bingcan Guo", "Shang Ma", "Ibrahim Khalilov", "Simret A Gebreegziabher", "Yanfang Ye", "Ziang Xiao", "Yaxing Yao", "Tianshi Li", "Toby Jia-Jun Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "48acf3279b854d45ec076e630e5a84601879fc9ff894c5d368adb59d03bd3de6", "submitted_at": "2025-04-15T15:21:09Z", "updated_at": "2025-04-15T15:21:09Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2504.11281v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2504.11281", "identifiers": {"arxiv": "2504.11281"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2504.11281"}], "name": "The Obvious Invisible Threat: LLM-Powered GUI Agents' Vulnerability to Fine-Print Injections", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-04-15", "relations": [], "review_status": "curated", "scope": "core", "summary": "Characterizes six forms of GUI-content injection and studies their effects on GUI agents and human users using adversarial webpages. The analysis examines privacy failures and the limits of human oversight.", "topics": ["safety-security", "human-in-the-loop", "evaluation"], "updated_at": "2025-04-15", "url": "https://arxiv.org/abs/2504.11281"}
{"aliases": [], "authors": ["Leo Boisvert", "Mihir Bansal", "Chandra Kiran Reddy Evuru", "Gabriel Huang", "Abhay Puri", "Avinandan Bose", "Maryam Fazel", "Quentin Cappart", "Jason Stanley", "Alexandre Lacoste", "Alexandre Drouin", "Krishnamurthy Dvijotham"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "c5bdf5fefa1ee31126d99de08236991df559896212f60e0b3058fcd8cc8a134b", "submitted_at": "2025-04-18T20:36:10Z", "updated_at": "2025-10-07T13:10:30Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2504.14064v3"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:08Z", "role": "primary", "sha256": "00d275cd5407cec847fa9d43fe73cfc4384c5570b68a5e058c97921799be0bec", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2302.01560%2C2303.16563%2C2305.15486%2C2305.16291%2C2305.17144%2C2306.00003%2C2306.00349%2C2306.17882%2C2310.01557%2C2310.08235%2C2310.08367%2C2311.05997%2C2311.17600%2C2312.02519%2C2312.11865%2C2402.01118%2C2402.10196%2C2402.11208%2C2403.10249%2C2403.12037%2C2404.06647%2C2404.12241%2C2404.13246%2C2407.15325%2C2409.00138%2C2410.14573%2C2410.18927%2C2410.19012%2C2411.09523%2C2411.10349%2C2411.13543%2C2412.01547%2C2501.05366%2C2501.09004%2C2502.07463%2C2502.14538%2C2503.02732%2C2503.08175%2C2503.11387%2C2504.10147%2C2504.11703%2C2504.14064%2C2504.15810%2C2504.16208%2C2504.20635%2C2504.21776%2C2505.12981%2C2505.13076%2C2506.01616%2C2506.13205&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:08Z", "role": "primary", "sha256": "00d275cd5407cec847fa9d43fe73cfc4384c5570b68a5e058c97921799be0bec", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2302.01560%2C2303.16563%2C2305.15486%2C2305.16291%2C2305.17144%2C2306.00003%2C2306.00349%2C2306.17882%2C2310.01557%2C2310.08235%2C2310.08367%2C2311.05997%2C2311.17600%2C2312.02519%2C2312.11865%2C2402.01118%2C2402.10196%2C2402.11208%2C2403.10249%2C2403.12037%2C2404.06647%2C2404.12241%2C2404.13246%2C2407.15325%2C2409.00138%2C2410.14573%2C2410.18927%2C2410.19012%2C2411.09523%2C2411.10349%2C2411.13543%2C2412.01547%2C2501.05366%2C2501.09004%2C2502.07463%2C2502.14538%2C2503.02732%2C2503.08175%2C2503.11387%2C2504.10147%2C2504.11703%2C2504.14064%2C2504.15810%2C2504.16208%2C2504.20635%2C2504.21776%2C2505.12981%2C2505.13076%2C2506.01616%2C2506.13205&max_results=50"}], "id": "arxiv:2504.14064", "identifiers": {"arxiv": "2504.14064"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:08Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2504.14064"}], "name": "DoomArena: A framework for Testing AI Agents Against Evolving Security Threats", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-04-18", "relations": [], "review_status": "curated", "scope": "core", "summary": "Presents a modular security testing framework that plugs into BrowserGym and tool-use benchmarks to evaluate agents under configurable threat models.", "topics": ["safety-security", "evaluation"], "updated_at": "2025-10-07", "url": "https://arxiv.org/abs/2504.14064"}
{"aliases": [], "authors": ["Chaoran Chen", "Zhiping Zhang", "Ibrahim Khalilov", "Bingcan Guo", "Simret A Gebreegziabher", "Yanfang Ye", "Ziang Xiao", "Yaxing Yao", "Tianshi Li", "Toby Jia-Jun Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "d2a55e347bc10e5571464924d9301ff2c32d5e9f1f990ca91725ec8c9f6da594", "submitted_at": "2025-04-24T20:51:20Z", "updated_at": "2025-06-05T12:40:53Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2504.17934v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2504.17934", "identifiers": {"arxiv": "2504.17934"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2504.17934"}], "name": "Toward a Human-Centered Evaluation Framework for Trustworthy LLM-Powered GUI Agents", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced. This is a position paper proposing an evaluation framework; the abstract does not report a new benchmark or empirical validation of the proposed framework.", "published_at": "2025-04-24", "relations": [], "review_status": "curated", "scope": "core", "summary": "Presents a position paper on privacy and security evaluation for GUI agents, identifying risk categories and challenges in involving human evaluators. It proposes risk-aware assessment and in-context consent as design directions.", "topics": ["evaluation", "safety-security", "human-in-the-loop"], "updated_at": "2025-06-05", "url": "https://arxiv.org/abs/2504.17934"}
{"aliases": [], "authors": ["Ivan Evtimov", "Arman Zharmagambetov", "Aaron Grattafiori", "Chuan Guo", "Kamalika Chaudhuri"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "3a104d9076797555795766749f6b4eeec9428d0efb9340f7e55cffecd5d727b0", "submitted_at": "2025-04-22T17:51:03Z", "updated_at": "2025-05-16T22:42:29Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2504.18575v3"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2504.18575", "identifiers": {"arxiv": "2504.18575"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2504.18575"}], "name": "WASP: Benchmarking Web Agent Security Against Prompt Injection Attacks", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-04-22", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces WASP, an end-to-end benchmark for prompt-injection security in web agents. Its evaluation separates partial attack progress from completion of the attacker’s full objective.", "topics": ["safety-security", "evaluation", "web-navigation"], "updated_at": "2025-05-16", "url": "https://arxiv.org/abs/2504.18575"}
{"aliases": [], "authors": ["Shuning Zhang", "Jingruo Chen", "Zhiqi Gao", "Jiajing Gao", "Xin Yi", "Hewu Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "63f4a505645fec16eac6b59923a6c1979c53376f19c6a4ea53c9a5d32a45e3d4", "submitted_at": "2025-05-15T00:42:51Z", "updated_at": "2025-05-16T07:01:20Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2505.09875v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2505.09875", "identifiers": {"arxiv": "2505.09875"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2505.09875"}], "name": "Characterizing Unintended Consequences in Human-GUI Agent Collaboration for Web Browsing", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-05-15", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies unintended consequences of collaborative web browsing with GUI agents through social-media analysis and interviews. It organizes observed failures, user impacts, and mitigation practices around planning, interface interaction, feedback, privacy, and trust.", "topics": ["human-in-the-loop", "evaluation", "safety-security"], "updated_at": "2025-05-16", "url": "https://arxiv.org/abs/2505.09875"}
{"aliases": [], "authors": ["Ada Chen", "Yongjiang Wu", "Junyuan Zhang", "Jingyu Xiao", "Shu Yang", "Jen-tse Huang", "Kun Wang", "Wenxuan Wang", "Shuai Wang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e8a0890cd57809ff5e3bede6dc2b851b42fad9307a5bfa0ddd744cc2ac70d0c2", "submitted_at": "2025-05-16T06:56:42Z", "updated_at": "2026-04-29T05:56:22Z", "version": 4, "versioned_url": "https://arxiv.org/abs/2505.10924v4"}}, "environments": ["cross-platform"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2505.10924", "identifiers": {"arxiv": "2505.10924"}, "kind": "survey", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2505.10924"}], "name": "A Survey on the Safety and Security Threats of Computer-Using Agents: JARVIS or Ultron?", "notes": "", "published_at": "2025-05-16", "relations": [], "review_status": "curated", "scope": "core", "summary": "Systematizes safety and security threats to computer-using agents, with taxonomies of threats and defensive strategies.", "topics": ["survey", "safety-security"], "updated_at": "2026-04-29", "url": "https://arxiv.org/abs/2505.10924"}
{"aliases": [], "authors": ["Zheng Wu", "Pengzhou Cheng", "Zongru Wu", "Lingzhong Dong", "Zhuosheng Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "ce9570eb6a52feb220cdd2eb4c7f37a2ea63082febe48888a2654c974138a82d", "submitted_at": "2025-05-19T08:29:05Z", "updated_at": "2025-11-22T15:07:27Z", "version": 4, "versioned_url": "https://arxiv.org/abs/2505.12842v4"}}, "environments": ["desktop", "web", "cross-platform"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2505.12842", "identifiers": {"arxiv": "2505.12842"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2505.12842"}], "name": "GEM: Gaussian Embedding Modeling for Out-of-Distribution Detection in GUI Agents", "notes": "The annotation was reviewed against the complete retained primary abstract. Experiments and released artifacts have not been independently reproduced.", "published_at": "2025-05-19", "relations": [], "review_status": "curated", "scope": "core", "summary": "GEM detects instructions outside a GUI agent’s capabilities by modeling distances in its input-embedding space with a Gaussian mixture. Experiments across mobile, computer and browser data also examine selective escalation to a cloud model when an input is detected as unfamiliar.", "topics": ["robustness", "safety-security", "evaluation", "efficiency"], "updated_at": "2025-11-22", "url": "https://arxiv.org/abs/2505.12842"}
{"aliases": [], "authors": ["Liangxuan Wu", "Chao Wang", "Tianming Liu", "Yanjie Zhao", "Haoyu Wang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "9657a66aebe46151ed56bf76131ffe45f52459c1cf0e000907113776c91259d9", "submitted_at": "2025-05-19T11:17:46Z", "updated_at": "2025-05-20T07:02:05Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2505.12981v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:08Z", "role": "primary", "sha256": "00d275cd5407cec847fa9d43fe73cfc4384c5570b68a5e058c97921799be0bec", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2302.01560%2C2303.16563%2C2305.15486%2C2305.16291%2C2305.17144%2C2306.00003%2C2306.00349%2C2306.17882%2C2310.01557%2C2310.08235%2C2310.08367%2C2311.05997%2C2311.17600%2C2312.02519%2C2312.11865%2C2402.01118%2C2402.10196%2C2402.11208%2C2403.10249%2C2403.12037%2C2404.06647%2C2404.12241%2C2404.13246%2C2407.15325%2C2409.00138%2C2410.14573%2C2410.18927%2C2410.19012%2C2411.09523%2C2411.10349%2C2411.13543%2C2412.01547%2C2501.05366%2C2501.09004%2C2502.07463%2C2502.14538%2C2503.02732%2C2503.08175%2C2503.11387%2C2504.10147%2C2504.11703%2C2504.14064%2C2504.15810%2C2504.16208%2C2504.20635%2C2504.21776%2C2505.12981%2C2505.13076%2C2506.01616%2C2506.13205&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:08Z", "role": "primary", "sha256": "00d275cd5407cec847fa9d43fe73cfc4384c5570b68a5e058c97921799be0bec", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2302.01560%2C2303.16563%2C2305.15486%2C2305.16291%2C2305.17144%2C2306.00003%2C2306.00349%2C2306.17882%2C2310.01557%2C2310.08235%2C2310.08367%2C2311.05997%2C2311.17600%2C2312.02519%2C2312.11865%2C2402.01118%2C2402.10196%2C2402.11208%2C2403.10249%2C2403.12037%2C2404.06647%2C2404.12241%2C2404.13246%2C2407.15325%2C2409.00138%2C2410.14573%2C2410.18927%2C2410.19012%2C2411.09523%2C2411.10349%2C2411.13543%2C2412.01547%2C2501.05366%2C2501.09004%2C2502.07463%2C2502.14538%2C2503.02732%2C2503.08175%2C2503.11387%2C2504.10147%2C2504.11703%2C2504.14064%2C2504.15810%2C2504.16208%2C2504.20635%2C2504.21776%2C2505.12981%2C2505.13076%2C2506.01616%2C2506.13205&max_results=50"}], "id": "arxiv:2505.12981", "identifiers": {"arxiv": "2505.12981"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:08Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2505.12981"}], "name": "From Assistants to Adversaries: Exploring the Security Risks of Mobile LLM Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-05-19", "relations": [], "review_status": "curated", "scope": "core", "summary": "Analyzes security vulnerabilities across OEM system-level, third-party, and research mobile LLM agents, covering GUI and cross-application attack surfaces.", "topics": ["safety-security", "mobile-navigation", "evaluation"], "updated_at": "2025-05-20", "url": "https://arxiv.org/abs/2505.12981"}
{"aliases": [], "authors": ["Mykyta Mudryi", "Markiyan Chaklosh", "Grzegorz Wójcik"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "709ee511ffa39e805fcc58abe066cf2cec6c1707a4a9c64d9cfd107bef7544af", "submitted_at": "2025-05-19T13:10:29Z", "updated_at": "2025-05-19T13:10:29Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2505.13076v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:08Z", "role": "primary", "sha256": "00d275cd5407cec847fa9d43fe73cfc4384c5570b68a5e058c97921799be0bec", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2302.01560%2C2303.16563%2C2305.15486%2C2305.16291%2C2305.17144%2C2306.00003%2C2306.00349%2C2306.17882%2C2310.01557%2C2310.08235%2C2310.08367%2C2311.05997%2C2311.17600%2C2312.02519%2C2312.11865%2C2402.01118%2C2402.10196%2C2402.11208%2C2403.10249%2C2403.12037%2C2404.06647%2C2404.12241%2C2404.13246%2C2407.15325%2C2409.00138%2C2410.14573%2C2410.18927%2C2410.19012%2C2411.09523%2C2411.10349%2C2411.13543%2C2412.01547%2C2501.05366%2C2501.09004%2C2502.07463%2C2502.14538%2C2503.02732%2C2503.08175%2C2503.11387%2C2504.10147%2C2504.11703%2C2504.14064%2C2504.15810%2C2504.16208%2C2504.20635%2C2504.21776%2C2505.12981%2C2505.13076%2C2506.01616%2C2506.13205&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:08Z", "role": "primary", "sha256": "00d275cd5407cec847fa9d43fe73cfc4384c5570b68a5e058c97921799be0bec", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2302.01560%2C2303.16563%2C2305.15486%2C2305.16291%2C2305.17144%2C2306.00003%2C2306.00349%2C2306.17882%2C2310.01557%2C2310.08235%2C2310.08367%2C2311.05997%2C2311.17600%2C2312.02519%2C2312.11865%2C2402.01118%2C2402.10196%2C2402.11208%2C2403.10249%2C2403.12037%2C2404.06647%2C2404.12241%2C2404.13246%2C2407.15325%2C2409.00138%2C2410.14573%2C2410.18927%2C2410.19012%2C2411.09523%2C2411.10349%2C2411.13543%2C2412.01547%2C2501.05366%2C2501.09004%2C2502.07463%2C2502.14538%2C2503.02732%2C2503.08175%2C2503.11387%2C2504.10147%2C2504.11703%2C2504.14064%2C2504.15810%2C2504.16208%2C2504.20635%2C2504.21776%2C2505.12981%2C2505.13076%2C2506.01616%2C2506.13205&max_results=50"}], "id": "arxiv:2505.13076", "identifiers": {"arxiv": "2505.13076"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:08Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2505.13076"}], "name": "The Hidden Dangers of Browsing AI Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-05-19", "relations": [], "review_status": "curated", "scope": "core", "summary": "Outlines an end-to-end threat model and empirical security evaluation across architectural layers of autonomous browser agents.", "topics": ["safety-security", "web-navigation", "evaluation"], "updated_at": "2025-05-19", "url": "https://arxiv.org/abs/2505.13076"}
{"aliases": [], "authors": ["Yijie Lu", "Manman Zhao", "Tianjie Ju", "Zihe Yan", "Xinbei Ma", "Yuan Guo", "Daizong Ding", "Gongshen Liu", "Zhuosheng Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "2c256d8b0d774efdeee896860f8c504c0d1593215508fdf0881d3fd3cdc3968c", "submitted_at": "2025-05-20T12:41:05Z", "updated_at": "2026-06-05T10:24:33Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2505.14289v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2505.14289", "identifiers": {"arxiv": "2505.14289"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2505.14289"}], "name": "EVA: Evolving Semantic Adversaries for Red-Teaming GUI Agents Against Environmental Injection Attacks", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-05-20", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces EVA, a red-teaming framework that evolves semantic payloads for GUI environmental-injection tests. It studies the contribution of semantic versus visual deception and distills discovered patterns into reusable evaluation rules.", "topics": ["safety-security", "evaluation"], "updated_at": "2026-06-05", "url": "https://arxiv.org/abs/2505.14289"}
{"aliases": [], "authors": ["Pengzhou Cheng", "Haowen Hu", "Zheng Wu", "Zongru Wu", "Tianjie Ju", "Zhuosheng Zhang", "Gongshen Liu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "815345ed44d70f40ce9784c212d269ffaaa476e5c2ea5cdbbaaeb98d22eba3cb", "submitted_at": "2025-05-20T14:29:18Z", "updated_at": "2026-06-15T06:42:51Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2505.14418v3"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2505.14418", "identifiers": {"arxiv": "2505.14418"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2505.14418"}], "name": "Hidden Ghost Hand: Unveiling Backdoor Vulnerabilities in MLLM-Powered Mobile GUI Agents", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-05-20", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies backdoor risks in mobile GUI agents through AgentGhost, which combines goal-level and interaction-history triggers during model training. The paper evaluates multiple mobile agents and a tailored defense.", "topics": ["safety-security", "mobile-navigation", "interaction-history"], "updated_at": "2026-06-15", "url": "https://arxiv.org/abs/2505.14418"}
{"aliases": [], "authors": ["Haowei Wang", "Junjie Wang", "Xiaojun Jia", "Rupeng Zhang", "Mingyang Li", "Zhe Liu", "Yang Liu", "Qing Wang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "f8aade17e67471c87f259d49aef43b3caf764b6ac14f99e1a4114f4a25dda3d6", "submitted_at": "2025-05-27T17:59:05Z", "updated_at": "2025-05-27T17:59:05Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2505.21499v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2505.21499", "identifiers": {"arxiv": "2505.21499"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2505.21499"}], "name": "AdInject: Real-World Black-Box Attacks on Web Agents via Advertising Delivery", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-05-27", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies advertising delivery as an indirect-injection channel for web agents under a black-box threat model. AdInject evaluates how static ad content shaped by website context can redirect browser actions.", "topics": ["safety-security", "web-navigation", "evaluation"], "updated_at": "2025-05-27", "url": "https://arxiv.org/abs/2505.21499"}
{"aliases": [], "authors": ["Zeyi Liao", "Jaylen Jones", "Linxi Jiang", "Yuting Ning", "Eric Fosler-Lussier", "Yu Su", "Zhiqiang Lin", "Huan Sun"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "7fe3bdb1357a375de0b1700a6eda7d5987f358bd47fddf3405c27272a3453f2f", "submitted_at": "2025-05-28T03:42:09Z", "updated_at": "2026-03-01T07:50:09Z", "version": 5, "versioned_url": "https://arxiv.org/abs/2505.21936v5"}}, "environments": ["web", "desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2505.21936", "identifiers": {"arxiv": "2505.21936"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2505.21936"}], "name": "RedTeamCUA: Realistic Adversarial Testing of Computer-Use Agents in Hybrid Web-OS Environments", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-05-28", "relations": [], "review_status": "curated", "scope": "core", "summary": "RedTeamCUA combines a virtual-machine OS environment with containerized websites for controlled prompt-injection tests spanning both interfaces. Its RTC-Bench contains 864 examples and can initialize agents at injection points to separate attack susceptibility from navigation failures.", "topics": ["safety-security", "evaluation", "sandbox"], "updated_at": "2026-03-01", "url": "https://arxiv.org/abs/2505.21936"}
{"aliases": [], "authors": ["Jingyi Yang", "Shuai Shao", "Dongrui Liu", "Jing Shao"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "bb94fdba775cb0fb54b312fdff9fa71afd3aefa56b812ff31ceccad33eb5f800", "submitted_at": "2025-05-31T16:04:59Z", "updated_at": "2025-06-20T01:24:06Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2506.00618v3"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2506.00618", "identifiers": {"arxiv": "2506.00618"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2506.00618"}], "name": "RiOSWorld: Benchmarking the Risk of Multimodal Computer-Use Agents", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-05-31", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces RiOSWorld, a computer-use risk benchmark separating user-originated and environmental threats. It evaluates both risky-goal intent and goal completion across web and other computer applications.", "topics": ["safety-security", "evaluation"], "updated_at": "2025-06-20", "url": "https://arxiv.org/abs/2506.00618"}
{"aliases": [], "authors": ["Xiao Yang", "Jiawei Chen", "Jun Luo", "Zhengwei Fang", "Yinpeng Dong", "Hang Su", "Jun Zhu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "1ec2467703bc5d0f97ef8d207289ed56f6d12b811688d20b7833eac06038757b", "submitted_at": "2025-06-02T12:56:27Z", "updated_at": "2025-06-02T12:56:27Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2506.01616v1"}}, "environments": ["web", "android", "cross-platform"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:08Z", "role": "primary", "sha256": "00d275cd5407cec847fa9d43fe73cfc4384c5570b68a5e058c97921799be0bec", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2302.01560%2C2303.16563%2C2305.15486%2C2305.16291%2C2305.17144%2C2306.00003%2C2306.00349%2C2306.17882%2C2310.01557%2C2310.08235%2C2310.08367%2C2311.05997%2C2311.17600%2C2312.02519%2C2312.11865%2C2402.01118%2C2402.10196%2C2402.11208%2C2403.10249%2C2403.12037%2C2404.06647%2C2404.12241%2C2404.13246%2C2407.15325%2C2409.00138%2C2410.14573%2C2410.18927%2C2410.19012%2C2411.09523%2C2411.10349%2C2411.13543%2C2412.01547%2C2501.05366%2C2501.09004%2C2502.07463%2C2502.14538%2C2503.02732%2C2503.08175%2C2503.11387%2C2504.10147%2C2504.11703%2C2504.14064%2C2504.15810%2C2504.16208%2C2504.20635%2C2504.21776%2C2505.12981%2C2505.13076%2C2506.01616%2C2506.13205&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:08Z", "role": "primary", "sha256": "00d275cd5407cec847fa9d43fe73cfc4384c5570b68a5e058c97921799be0bec", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2302.01560%2C2303.16563%2C2305.15486%2C2305.16291%2C2305.17144%2C2306.00003%2C2306.00349%2C2306.17882%2C2310.01557%2C2310.08235%2C2310.08367%2C2311.05997%2C2311.17600%2C2312.02519%2C2312.11865%2C2402.01118%2C2402.10196%2C2402.11208%2C2403.10249%2C2403.12037%2C2404.06647%2C2404.12241%2C2404.13246%2C2407.15325%2C2409.00138%2C2410.14573%2C2410.18927%2C2410.19012%2C2411.09523%2C2411.10349%2C2411.13543%2C2412.01547%2C2501.05366%2C2501.09004%2C2502.07463%2C2502.14538%2C2503.02732%2C2503.08175%2C2503.11387%2C2504.10147%2C2504.11703%2C2504.14064%2C2504.15810%2C2504.16208%2C2504.20635%2C2504.21776%2C2505.12981%2C2505.13076%2C2506.01616%2C2506.13205&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:59:55Z", "role": "primary", "sha256": "d69da19edaed185ad3071a97db3bb3fcfbdb14c0b514b7d27794b39fbe8e2541", "supports": ["environments", "notes"], "url": "https://raw.githubusercontent.com/thu-ml/MLA-Trust/HEAD/README.md"}], "id": "arxiv:2506.01616", "identifiers": {"arxiv": "2506.01616"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:08Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2506.01616"}], "name": "MLA-Trust: Benchmarking Trustworthiness of Multimodal LLM Agents in GUI Environments", "notes": "The contribution was reviewed from the retained paper abstract; Android setup is supported separately by the official project README. No benchmark results were independently reproduced.", "published_at": "2025-06-02", "relations": [], "review_status": "curated", "scope": "core", "summary": "Benchmarks truthfulness, controllability, safety, and privacy of multimodal LLM agents across interactive web and mobile GUI tasks.", "topics": ["safety-security", "evaluation", "robustness"], "updated_at": "2025-06-02", "url": "https://arxiv.org/abs/2506.01616"}
{"aliases": [], "authors": ["Tri Cao", "Bennett Lim", "Yue Liu", "Yuan Sui", "Yuexin Li", "Shumin Deng", "Lin Lu", "Nay Oo", "Shuicheng Yan", "Bryan Hooi"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "3cd988217d1e3371bec827b472abf46c205ac407b5b60b7b581d7105f86e8bc5", "submitted_at": "2025-06-03T05:21:50Z", "updated_at": "2026-03-01T16:12:40Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2506.02456v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2506.02456", "identifiers": {"arxiv": "2506.02456"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2506.02456"}], "name": "VPI-Bench: Visual Prompt Injection Attacks for Computer-Use Agents", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-06-03", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces VPI-Bench, interactive web-platform variants containing visually embedded malicious instructions. It evaluates computer-use and browser-use agents under these threats and examines system-prompt defenses.", "topics": ["safety-security", "evaluation"], "updated_at": "2026-03-01", "url": "https://arxiv.org/abs/2506.02456"}
{"aliases": [], "authors": ["Thong Q. Nguyen", "Shubhang Desai", "Raja Hasnain Anwar", "Firoz Shaik", "Vishwas Suryanarayanan", "Vishal Chowdhary"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "401ca18cadbf38045fa20985354988571b69e2280b1a090aff0e128cdf3c70e0", "submitted_at": "2025-06-03T07:25:49Z", "updated_at": "2025-08-07T21:48:01Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2506.02539v3"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2506.02539", "identifiers": {"arxiv": "2506.02539"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2506.02539"}], "name": "VerificAgent: Domain-Specific Memory Verification for Scalable Oversight of Aligned Computer-Use Agents", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-06-03", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces VerificAgent, which grows domain-specific memories from task trajectories and subjects them to human fact-checking before deployment. Its OSWorld evaluation treats persistent memory as an auditable constraint on later agent behavior.", "topics": ["memory", "human-in-the-loop", "safety-security"], "updated_at": "2025-08-07", "url": "https://arxiv.org/abs/2506.02539"}
{"aliases": [], "authors": ["Pei Yang", "Hai Ci", "Mike Zheng Shou"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "8a4c980694819dc5eb4ff322e6a703cccfe0b1e45efb60f22b120592d03edf36", "submitted_at": "2025-06-04T16:26:56Z", "updated_at": "2025-10-18T12:11:31Z", "version": 4, "versioned_url": "https://arxiv.org/abs/2506.04135v4"}}, "environments": ["macos", "desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2506.04135", "identifiers": {"arxiv": "2506.04135"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2506.04135"}], "name": "macOSWorld: A Multilingual Interactive Benchmark for GUI Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-06-04", "relations": [], "review_status": "curated", "scope": "core", "summary": "macOSWorld provides 202 interactive tasks across 30 applications, with instructions and operating-system interfaces in five languages. It evaluates GUI agents on macOS workflows and includes a separate subset for deception-attack testing.", "topics": ["evaluation", "workflow", "safety-security"], "updated_at": "2025-10-18", "url": "https://arxiv.org/abs/2506.04135"}
{"aliases": [], "authors": ["Xuan Wang", "Siyuan Liang", "Zhe Liu", "Yi Yu", "Aishan Liu", "Yuliang Lu", "Xitong Gao", "Ee-Chien Chang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "390304ce515886fa15fb8cd74d8b2df161a75b9adc0f4d2fad1f65976c6a6ce7", "submitted_at": "2025-06-16T08:09:32Z", "updated_at": "2025-09-05T14:19:03Z", "version": 6, "versioned_url": "https://arxiv.org/abs/2506.13205v6"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:08Z", "role": "primary", "sha256": "00d275cd5407cec847fa9d43fe73cfc4384c5570b68a5e058c97921799be0bec", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2302.01560%2C2303.16563%2C2305.15486%2C2305.16291%2C2305.17144%2C2306.00003%2C2306.00349%2C2306.17882%2C2310.01557%2C2310.08235%2C2310.08367%2C2311.05997%2C2311.17600%2C2312.02519%2C2312.11865%2C2402.01118%2C2402.10196%2C2402.11208%2C2403.10249%2C2403.12037%2C2404.06647%2C2404.12241%2C2404.13246%2C2407.15325%2C2409.00138%2C2410.14573%2C2410.18927%2C2410.19012%2C2411.09523%2C2411.10349%2C2411.13543%2C2412.01547%2C2501.05366%2C2501.09004%2C2502.07463%2C2502.14538%2C2503.02732%2C2503.08175%2C2503.11387%2C2504.10147%2C2504.11703%2C2504.14064%2C2504.15810%2C2504.16208%2C2504.20635%2C2504.21776%2C2505.12981%2C2505.13076%2C2506.01616%2C2506.13205&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:08Z", "role": "primary", "sha256": "00d275cd5407cec847fa9d43fe73cfc4384c5570b68a5e058c97921799be0bec", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2302.01560%2C2303.16563%2C2305.15486%2C2305.16291%2C2305.17144%2C2306.00003%2C2306.00349%2C2306.17882%2C2310.01557%2C2310.08235%2C2310.08367%2C2311.05997%2C2311.17600%2C2312.02519%2C2312.11865%2C2402.01118%2C2402.10196%2C2402.11208%2C2403.10249%2C2403.12037%2C2404.06647%2C2404.12241%2C2404.13246%2C2407.15325%2C2409.00138%2C2410.14573%2C2410.18927%2C2410.19012%2C2411.09523%2C2411.10349%2C2411.13543%2C2412.01547%2C2501.05366%2C2501.09004%2C2502.07463%2C2502.14538%2C2503.02732%2C2503.08175%2C2503.11387%2C2504.10147%2C2504.11703%2C2504.14064%2C2504.15810%2C2504.16208%2C2504.20635%2C2504.21776%2C2505.12981%2C2505.13076%2C2506.01616%2C2506.13205&max_results=50"}], "id": "arxiv:2506.13205", "identifiers": {"arxiv": "2506.13205"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:08Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2506.13205"}], "name": "Poison Once, Control Anywhere: Clean-Text Visual Backdoors in VLM-based Mobile Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-06-16", "relations": [], "review_status": "curated", "scope": "core", "summary": "Demonstrates clean-text visual backdoor attacks (VIBMA) that compromise fine-tuned VLM mobile agents through visual triggers while keeping text instructions intact.", "topics": ["safety-security", "mobile-navigation", "robustness"], "updated_at": "2025-09-05", "url": "https://arxiv.org/abs/2506.13205"}
{"aliases": [], "authors": ["Thomas Kuntz", "Agatha Duzan", "Hao Zhao", "Francesco Croce", "Zico Kolter", "Nicolas Flammarion", "Maksym Andriushchenko"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "f551b861012fc4eddbf010a619bc65e77b524b05958761b738a88bcae377e9a9", "submitted_at": "2025-06-17T17:59:31Z", "updated_at": "2025-10-29T10:34:04Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2506.14866v2"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2506.14866", "identifiers": {"arxiv": "2506.14866"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2506.14866"}], "name": "OS-Harm: A Benchmark for Measuring Safety of Computer Use Agents", "notes": "Summary and topic tags reviewed against the primary abstract. Reported experiments, task counts, and artifact availability are not independently reproduced.", "published_at": "2025-06-17", "relations": [], "review_status": "curated", "scope": "core", "summary": "Extends OSWorld with tasks covering user misuse, prompt injection, and model misbehavior, with assessment of both task success and safety.", "topics": ["safety-security", "evaluation"], "updated_at": "2025-10-29", "url": "https://arxiv.org/abs/2506.14866"}
{"aliases": [], "authors": ["Atharv Singh Patlan", "Ashwin Hebbar", "Pramod Viswanath", "Prateek Mittal"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "2d346389013d465f578417a7fca80ee527f5e7b582e93eb73dae148f11479971", "submitted_at": "2025-06-18T14:29:02Z", "updated_at": "2025-06-18T14:29:02Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2506.17318v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2506.17318", "identifiers": {"arxiv": "2506.17318"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2506.17318"}], "name": "Context manipulation attacks : Web agents are susceptible to corrupted memory", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-06-18", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies context manipulation in web agents by altering stored task representations rather than webpage prompts alone. Evaluation on Browser-use and Agent-E examines memory-directed attacks and the implications for secure context handling.", "topics": ["safety-security", "memory", "evaluation"], "updated_at": "2025-06-18", "url": "https://arxiv.org/abs/2506.17318"}
{"aliases": [], "authors": ["Siyuan Liang", "Tianmeng Fang", "Zhe Liu", "Aishan Liu", "Yan Xiao", "Jinyuan He", "Ee-Chien Chang", "Xiaochun Cao"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "78dd523ad6123fe1ee5164685a3853719c9a4442adac8661050f147ff6a9f619", "submitted_at": "2025-07-01T15:10:00Z", "updated_at": "2025-07-01T15:10:00Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2507.00841v1"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:01:16.030926+00:00", "role": "primary", "sha256": "4dcd1ba66ba21c0f2cb2aeb00c8bcab7c0cee4da37ff2fa6c16b45066aaa9db0", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22+OR+ti%3A%22mobile+agent%22+OR+ti%3A%22UI+agent%22+OR+ti%3A%22desktop+agent%22+OR+ti%3A%22screen+grounding%22%29+AND+submittedDate%3A%5B200001010000+TO+202610180000%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:01:16.030926+00:00", "role": "primary", "sha256": "4dcd1ba66ba21c0f2cb2aeb00c8bcab7c0cee4da37ff2fa6c16b45066aaa9db0", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22+OR+ti%3A%22mobile+agent%22+OR+ti%3A%22UI+agent%22+OR+ti%3A%22desktop+agent%22+OR+ti%3A%22screen+grounding%22%29+AND+submittedDate%3A%5B200001010000+TO+202610180000%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:10:56.912207+00:00", "role": "primary", "sha256": "a25da6700a238167d5e9915cf6fb3a4bcd910cefc300f4be6eed56271be26f00", "supports": ["summary", "scope", "environments", "notes"], "url": "https://arxiv.org/pdf/2507.00841v1"}], "id": "arxiv:2507.00841", "identifiers": {"arxiv": "2507.00841"}, "kind": "paper", "last_verified_at": "2026-10-18T10:01:16.030926+00:00", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2507.00841"}], "name": "SafeMobile: Chain-level Jailbreak Detection and Automated Evaluation for Multimodal Mobile Agents", "notes": "The reviewed evaluation uses Android emulation and a specific MobileSafetyBench-derived setup. Its judge and score threshold are evaluator choices, not universal safety measurements, and its task population is not the whole later MobileSafetyBench corpus. No results were reproduced. Basis: complete retained abstract and source-reviewed sections of the exact-version paper (II.B Defense objective; III Chain-level guard and GPTJudge; IV.B Evaluation setup; V.D Judge validity); not an entire-paper reading or reproduction claim.", "published_at": "2025-07-01", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies chain-level risk detection for mobile multimodal agents and a language-model-assisted evaluator, assessing safety interventions together with functional preservation.", "topics": ["safety-security", "evaluation", "mobile-navigation"], "updated_at": "2025-07-01", "url": "https://arxiv.org/abs/2507.00841"}
{"aliases": [], "authors": ["Guohong Liu", "Jialei Ye", "Jiacheng Liu", "Yuanchun Li", "Wei Liu", "Pengzhi Gao", "Jian Luan", "Yunxin Liu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "85ebf8b3b5916e66e4b4221c73bb29519156c41e8d1695bcc46d275dd50cf58d", "submitted_at": "2025-07-06T03:31:36Z", "updated_at": "2026-04-14T14:47:25Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2507.04227v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2507.04227", "identifiers": {"arxiv": "2507.04227"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2507.04227"}], "name": "Mobile GUI Agents under Real-world Threats: Are We There Yet?", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-07-06", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces an app-content instrumentation framework and complementary interactive and static test suites for mobile GUI agents. It studies how untrusted third-party content changes task execution and state understanding.", "topics": ["mobile-navigation", "safety-security", "evaluation"], "updated_at": "2026-04-14", "url": "https://arxiv.org/abs/2507.04227"}
{"aliases": [], "authors": ["Daniel Jones", "Giorgio Severi", "Martin Pouliot", "Gary Lopez", "Joris de Gruyter", "Santiago Zanella-Beguelin", "Justin Song", "Blake Bullwinkel", "Pamela Cortez", "Amanda Minnich"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "3ad2f299303076e3474cf247cf25d20ba63b1f1aa8601ab74243c00896bb56e4", "submitted_at": "2025-07-07T19:50:21Z", "updated_at": "2025-07-07T19:50:21Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2507.05445v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2507.05445", "identifiers": {"arxiv": "2507.05445"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2507.05445"}], "name": "A Systematization of Security Vulnerabilities in Computer Use Agents", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced. This is a threat analysis with case studies; the annotation does not imply a standardized public benchmark or independently validated defenses.", "published_at": "2025-07-07", "relations": [], "review_status": "curated", "scope": "core", "summary": "Analyzes security boundaries in computer-use agents through a risk taxonomy and concrete adversarial case studies. It links observed failures to input provenance, interface-action binding, memory, and delegation, and proposes CUA-specific evaluation principles.", "topics": ["safety-security", "evaluation"], "updated_at": "2025-07-07", "url": "https://arxiv.org/abs/2507.05445"}
{"aliases": [], "authors": ["Sanidhya Vijayvargiya", "Aditya Bharat Soni", "Xuhui Zhou", "Zora Zhiruo Wang", "Nouha Dziri", "Graham Neubig", "Maarten Sap"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "db56d120a7f04f9181a107bcd75c61a3eb3ac36939e2a0517d622332c70f3eac", "submitted_at": "2025-07-08T16:18:54Z", "updated_at": "2026-02-16T22:34:53Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2507.06134v2"}}, "environments": ["web", "desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T07:54:03Z", "role": "primary", "sha256": "13d5969845070cd2877dc51ae5300c2d623e5daef00c8a1552ffa46af3f721a5", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2401.05507%2C2406.07275%2C2409.16427%2C2410.00079%2C2503.07701%2C2503.22738%2C2505.15859%2C2505.17735%2C2505.23596%2C2507.05257%2C2507.06134%2C2507.21953%2C2507.23336%2C2508.06960%2C2508.08997%2C2509.02473%2C2509.23694%2C2510.09116%2C2510.14438%2C2511.18423%2C2512.12818%2C2512.13564%2C2601.10758%2C2601.12449%2C2601.21841%2C2603.01152%2C2603.19993%2C2603.20576%2C2604.04503%2C2604.17562%2C2604.23941%2C2605.16265%2C2606.06448%2C2606.06462%2C2606.29193%2C2607.01647%2C2607.13157&max_results=37"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T07:54:03Z", "role": "primary", "sha256": "13d5969845070cd2877dc51ae5300c2d623e5daef00c8a1552ffa46af3f721a5", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2401.05507%2C2406.07275%2C2409.16427%2C2410.00079%2C2503.07701%2C2503.22738%2C2505.15859%2C2505.17735%2C2505.23596%2C2507.05257%2C2507.06134%2C2507.21953%2C2507.23336%2C2508.06960%2C2508.08997%2C2509.02473%2C2509.23694%2C2510.09116%2C2510.14438%2C2511.18423%2C2512.12818%2C2512.13564%2C2601.10758%2C2601.12449%2C2601.21841%2C2603.01152%2C2603.19993%2C2603.20576%2C2604.04503%2C2604.17562%2C2604.23941%2C2605.16265%2C2606.06448%2C2606.06462%2C2606.29193%2C2607.01647%2C2607.13157&max_results=37"}], "id": "arxiv:2507.06134", "identifiers": {"arxiv": "2507.06134"}, "kind": "paper", "last_verified_at": "2026-10-16T07:54:03Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2507.06134"}], "name": "OpenAgentSafety: A Comprehensive Framework for Evaluating Real-World AI Agent Safety", "notes": "Editorial summary based on the primary paper abstract; experiments and released artifacts have not been independently reproduced.", "published_at": "2025-07-08", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates agent safety across real browser, code, filesystem, shell, and messaging tools using rule-based and model-based assessments.", "topics": ["safety-security", "evaluation"], "updated_at": "2026-02-16", "url": "https://arxiv.org/abs/2507.06134"}
{"aliases": [], "authors": ["Ziang Ye", "Yang Zhang", "Wentao Shi", "Xiaoyu You", "Fuli Feng", "Tat-Seng Chua"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "010a3a54ba37417c97e52434211375447418dd2e7ce1da6c339532534f0a920e", "submitted_at": "2025-07-09T14:36:00Z", "updated_at": "2025-09-24T14:33:17Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2507.06899v2"}}, "environments": ["web", "desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2507.06899", "identifiers": {"arxiv": "2507.06899"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2507.06899"}], "name": "VisualTrap: A Stealthy Backdoor Attack on GUI Agents via Visual Grounding Manipulation", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-07-09", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies backdoors that redirect GUI visual grounding even when the task plan is correct. VisualTrap evaluates whether poisoned grounding training transfers to downstream tasks and across web, mobile, and desktop interface settings.", "topics": ["safety-security", "grounding", "evaluation"], "updated_at": "2025-09-24", "url": "https://arxiv.org/abs/2507.06899"}
{"aliases": [], "authors": ["Zihe Yan", "Jiaping Gui", "Zhuosheng Zhang", "Gongshen Liu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "5db0008a053357b6dba38b55f29642ddbcbf38761c64684fd1725183ff8e56d3", "submitted_at": "2025-07-13T08:36:09Z", "updated_at": "2026-04-07T11:46:55Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2507.10610v3"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2507.10610", "identifiers": {"arxiv": "2507.10610"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2507.10610"}], "name": "LaSM: Layer-wise Scaling Mechanism for Defending Pop-up Attack on GUI Agents", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-07-13", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces LaSM, a training-free defense that selectively scales attention and MLP components to improve alignment with task-relevant screen regions. It evaluates the approach against pop-up-based GUI injection attacks.", "topics": ["safety-security", "grounding", "training-free"], "updated_at": "2026-04-07", "url": "https://arxiv.org/abs/2507.10610"}
{"aliases": [], "authors": ["Joel Nitu", "Heidrun Mühle", "Andreas Stöckl"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "9fe2d9d877a0666d46a065bd8b673c282c879ef0a2620e3b267793a91858b519", "submitted_at": "2025-07-17T07:10:27Z", "updated_at": "2025-11-12T13:34:24Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2507.12844v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2507.12844", "identifiers": {"arxiv": "2507.12844"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2507.12844"}], "name": "Machine-Readable Ads: Accessibility and Trust Patterns for AI Web Agents interacting with Online Advertisements", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2025-07-17", "relations": [], "review_status": "curated", "scope": "core", "summary": "Compares DOM-based and pixel-based web agents on a cloned news site containing advertisements, consent dialogs and paywalls. The study examines how interface presentation affects ad interaction and consequential subscription choices.", "topics": ["evaluation", "ui-design", "safety-security"], "updated_at": "2025-11-12", "url": "https://arxiv.org/abs/2507.12844"}
{"aliases": [], "authors": ["Boyuan Zheng", "Zeyi Liao", "Scott Salisbury", "Zeyuan Liu", "Michael Lin", "Qinyuan Zheng", "Zifan Wang", "Xiang Deng", "Dawn Song", "Huan Sun", "Yu Su"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "a8f082fd40174e01f40e97d64f7aa3d90e5f76e831df8f93afdc21aa2e35bfd5", "submitted_at": "2025-07-18T18:06:27Z", "updated_at": "2025-07-18T18:06:27Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2507.14293v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2507.14293", "identifiers": {"arxiv": "2507.14293"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2507.14293"}], "name": "WebGuard: Building a Generalizable Guardrail for Web Agents", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-07-18", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces WebGuard, a human-annotated dataset for predicting outcomes and risk levels of state-changing website actions. It provides generalization splits and evaluates specialized guardrail models.", "topics": ["safety-security", "reward-models", "evaluation"], "updated_at": "2025-07-18", "url": "https://arxiv.org/abs/2507.14293"}
{"aliases": [], "authors": ["Sam Johnson", "Viet Pham", "Thai Le"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "79963ac401384f0c23ecc65d58e616daacc93103456ceb39241dcd6d71c6591c", "submitted_at": "2025-07-20T03:10:13Z", "updated_at": "2025-07-20T03:10:13Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2507.14799v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2507.14799", "identifiers": {"arxiv": "2507.14799"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2507.14799"}], "name": "Manipulating LLM Web Agents with Indirect Prompt Injection Attack via HTML Accessibility Tree", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-07-20", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies indirect prompt injection through HTML content consumed as accessibility trees. The evaluation tests how manipulated page observations redirect browser-agent actions on real websites.", "topics": ["safety-security", "web-navigation", "evaluation"], "updated_at": "2025-07-20", "url": "https://arxiv.org/abs/2507.14799"}
{"aliases": [], "authors": ["Hussein Mozannar", "Gagan Bansal", "Cheng Tan", "Adam Fourney", "Victor Dibia", "Jingya Chen", "Jack Gerrits", "Tyler Payne", "Matheus Kunzler Maldaner", "Madeleine Grunde-McLaughlin", "Eric Zhu", "Griffin Bassman", "Jacob Alber", "Peter Chang", "Ricky Loynd", "Friederike Niedtner", "Ece Kamar", "Maya Murad", "Rafah Hosn", "Saleema Amershi"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "c9416ede89036d640412496b5f0def218b69271c5f2449689d4e4afacb43c556", "submitted_at": "2025-07-30T03:49:14Z", "updated_at": "2025-07-30T03:49:14Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2507.22358v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:31:49Z", "role": "primary", "sha256": "c208e36ae75cef4b32cfcb2f8182dd4aaaa218e3119bb3d4a2c20a5832f8b1ee", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2506.10387%2C2506.14866%2C2506.20332%2C2507.00008%2C2507.02592%2C2507.03730%2C2507.04227%2C2507.05445%2C2507.05673%2C2507.05720%2C2507.05791%2C2507.14293%2C2507.14799%2C2507.15846%2C2507.16853%2C2507.19132%2C2507.19478%2C2507.21071%2C2507.22025%2C2507.22358%2C2507.23779%2C2508.00935%2C2508.01330%2C2508.03700%2C2508.03923%2C2508.04010%2C2508.04026%2C2508.04482%2C2508.04700%2C2508.05615%2C2508.05731%2C2508.09057%2C2508.09123%2C2508.13167%2C2508.14040%2C2508.15144%2C2508.18689%2C2508.19679%2C2508.20096%2C2509.02544%2C2509.03891%2C2509.06477%2C2509.10723%2C2509.11250%2C2509.11543%2C2509.15221%2C2509.15566%2C2509.15738%2C2509.18119%2C2510.01354&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:31:49Z", "role": "primary", "sha256": "c208e36ae75cef4b32cfcb2f8182dd4aaaa218e3119bb3d4a2c20a5832f8b1ee", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2506.10387%2C2506.14866%2C2506.20332%2C2507.00008%2C2507.02592%2C2507.03730%2C2507.04227%2C2507.05445%2C2507.05673%2C2507.05720%2C2507.05791%2C2507.14293%2C2507.14799%2C2507.15846%2C2507.16853%2C2507.19132%2C2507.19478%2C2507.21071%2C2507.22025%2C2507.22358%2C2507.23779%2C2508.00935%2C2508.01330%2C2508.03700%2C2508.03923%2C2508.04010%2C2508.04026%2C2508.04482%2C2508.04700%2C2508.05615%2C2508.05731%2C2508.09057%2C2508.09123%2C2508.13167%2C2508.14040%2C2508.15144%2C2508.18689%2C2508.19679%2C2508.20096%2C2509.02544%2C2509.03891%2C2509.06477%2C2509.10723%2C2509.11250%2C2509.11543%2C2509.15221%2C2509.15566%2C2509.15738%2C2509.18119%2C2510.01354&max_results=50"}], "id": "arxiv:2507.22358", "identifiers": {"arxiv": "2507.22358"}, "kind": "paper", "last_verified_at": "2026-10-16T06:31:49Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2507.22358"}], "name": "Magentic-UI: Towards Human-in-the-loop Agentic Systems", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-07-30", "relations": [], "review_status": "curated", "scope": "core", "summary": "Magentic-UI combines browser interaction, code execution, file operations, and MCP tools in an interface for human-agent collaboration. It studies collaborative planning and task execution, action guards, and persistent memory through benchmark tasks, simulated users, user studies, and safety assessments.", "topics": ["human-in-the-loop", "tool-integration", "web-navigation", "safety-security"], "updated_at": "2025-07-30", "url": "https://arxiv.org/abs/2507.22358"}
{"aliases": [], "authors": ["Zeyu Huang", "Juyuan Wang", "Longfeng Chen", "Boyi Xiao", "Leng Cai", "Yawen Zeng", "Jin Xu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "d4efa329c3b106219c9b25f675be0c561743085ac999d2964458a5f22ff6e62f", "submitted_at": "2025-08-12T16:18:30Z", "updated_at": "2025-08-14T16:36:45Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2508.09057v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:31:49Z", "role": "primary", "sha256": "c208e36ae75cef4b32cfcb2f8182dd4aaaa218e3119bb3d4a2c20a5832f8b1ee", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2506.10387%2C2506.14866%2C2506.20332%2C2507.00008%2C2507.02592%2C2507.03730%2C2507.04227%2C2507.05445%2C2507.05673%2C2507.05720%2C2507.05791%2C2507.14293%2C2507.14799%2C2507.15846%2C2507.16853%2C2507.19132%2C2507.19478%2C2507.21071%2C2507.22025%2C2507.22358%2C2507.23779%2C2508.00935%2C2508.01330%2C2508.03700%2C2508.03923%2C2508.04010%2C2508.04026%2C2508.04482%2C2508.04700%2C2508.05615%2C2508.05731%2C2508.09057%2C2508.09123%2C2508.13167%2C2508.14040%2C2508.15144%2C2508.18689%2C2508.19679%2C2508.20096%2C2509.02544%2C2509.03891%2C2509.06477%2C2509.10723%2C2509.11250%2C2509.11543%2C2509.15221%2C2509.15566%2C2509.15738%2C2509.18119%2C2510.01354&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:31:49Z", "role": "primary", "sha256": "c208e36ae75cef4b32cfcb2f8182dd4aaaa218e3119bb3d4a2c20a5832f8b1ee", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?id_list=2506.10387%2C2506.14866%2C2506.20332%2C2507.00008%2C2507.02592%2C2507.03730%2C2507.04227%2C2507.05445%2C2507.05673%2C2507.05720%2C2507.05791%2C2507.14293%2C2507.14799%2C2507.15846%2C2507.16853%2C2507.19132%2C2507.19478%2C2507.21071%2C2507.22025%2C2507.22358%2C2507.23779%2C2508.00935%2C2508.01330%2C2508.03700%2C2508.03923%2C2508.04010%2C2508.04026%2C2508.04482%2C2508.04700%2C2508.05615%2C2508.05731%2C2508.09057%2C2508.09123%2C2508.13167%2C2508.14040%2C2508.15144%2C2508.18689%2C2508.19679%2C2508.20096%2C2509.02544%2C2509.03891%2C2509.06477%2C2509.10723%2C2509.11250%2C2509.11543%2C2509.15221%2C2509.15566%2C2509.15738%2C2509.18119%2C2510.01354&max_results=50"}], "id": "arxiv:2508.09057", "identifiers": {"arxiv": "2508.09057"}, "kind": "paper", "last_verified_at": "2026-10-16T06:31:49Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2508.09057"}], "name": "MVISU-Bench: Benchmarking Mobile Agents for Real-World Tasks by Multi-App, Vague, Interactive, Single-App and Unethical Instructions", "notes": "The annotation was reviewed against the complete retained primary abstract. Experiments and released artifacts have not been independently reproduced.", "published_at": "2025-08-12", "relations": [], "review_status": "curated", "scope": "core", "summary": "MVISU-Bench evaluates bilingual mobile tasks involving cross-app work, vague requests, user interaction and unethical instructions alongside ordinary single-app tasks. Its Aider prompting module is designed to clarify intent and mitigate risky execution within agent workflows.", "topics": ["evaluation", "mobile-navigation", "human-in-the-loop", "safety-security"], "updated_at": "2025-08-14", "url": "https://arxiv.org/abs/2508.09057"}
{"aliases": [], "authors": ["Xianren Zhang", "Shreyas Prasad", "Di Wang", "Qiuhai Zeng", "Suhang Wang", "Wenbo Yan", "Mat Hans"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "6c8336a707723227a9d651a3e9347e45907f28ed7421b577761ed57f7d70593d", "submitted_at": "2025-08-18T21:58:43Z", "updated_at": "2026-04-21T00:38:54Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2508.15832v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2508.15832", "identifiers": {"arxiv": "2508.15832"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2508.15832"}], "name": "A Functionality-Grounded Benchmark for Evaluating Web Agents in E-commerce Domains", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-08-18", "relations": [], "review_status": "curated", "scope": "core", "summary": "Amazon-Bench generates functionality-specific web tasks from page content and interactive controls, covering operations beyond product search. Its evaluator checks task completion and unintended changes to account state, such as address or purchase settings.", "topics": ["evaluation", "safety-security", "data-generation", "web-navigation"], "updated_at": "2026-04-21", "url": "https://arxiv.org/abs/2508.15832"}
{"aliases": [], "authors": ["Zhixin Lin", "Jungang Li", "Shidong Pan", "Yibo Shi", "Yue Yao", "Dongliang Xu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "815c8c1fa5993d2ef58798abd43a2b8945c476c27c777d30c1cbfacadd87af84", "submitted_at": "2025-08-27T00:41:28Z", "updated_at": "2025-09-03T07:48:34Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2508.19493v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2508.19493", "identifiers": {"arxiv": "2508.19493"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2508.19493"}], "name": "Mind the Third Eye! Benchmarking Privacy Awareness in MLLM-powered Smartphone Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-08-27", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces SAPA-Bench with 7,138 smartphone scenarios to evaluate whether MLLM-powered mobile agents detect and protect sensitive personal information on screen.", "topics": ["safety-security", "mobile-navigation", "evaluation"], "updated_at": "2025-09-03", "url": "https://arxiv.org/abs/2508.19493"}
{"aliases": [], "authors": ["Abhinav Kumar", "Jaechul Roh", "Ali Naseh", "Amir Houmansadr", "Eugene Bagdasarian"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "0f6ff68c59cede9cf59c039416a87ea39b926e3a3179b9f81af9354767befb93", "submitted_at": "2025-09-01T16:56:16Z", "updated_at": "2025-09-01T16:56:16Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2509.01619v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2509.01619", "identifiers": {"arxiv": "2509.01619"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2509.01619"}], "name": "Throttling Web Agents Using Reasoning Gates", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced. The contribution is an access-cost mechanism spanning websites and MCP servers, not a GUI navigation policy.", "published_at": "2025-09-01", "relations": [], "review_status": "curated", "scope": "adjacent", "summary": "Introduces reasoning gates that require agents to solve generated, verifiable puzzles before receiving access to resources. The throttling approach is deployed on a custom website and MCP servers and evaluated with web agents.", "topics": ["safety-security", "evaluation", "tool-integration"], "updated_at": "2025-09-01", "url": "https://arxiv.org/abs/2509.01619"}
{"aliases": [], "authors": ["Zheng Wu", "Heyuan Huang", "Xingyu Lou", "Xiangmou Qu", "Pengzhou Cheng", "Zongru Wu", "Weiwen Liu", "Weinan Zhang", "Jun Wang", "Zhaoxiang Wang", "Zhuosheng Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "3bd6b323a376d7606b8c5ec0615b992f47a388cbd3604191241f6a0a9b2cce6e", "submitted_at": "2025-09-09T09:46:01Z", "updated_at": "2026-04-03T02:36:28Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2509.07553v3"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-17T07:57:56Z", "role": "primary", "sha256": "3429cc2fc7737050cac964c976293fc4793e52c5913636c1f67d04243fbefb6a", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.20410%2C2508.20890%2C2508.21456%2C2509.00088%2C2509.00531%2C2509.00969%2C2509.01684%2C2509.03536%2C2509.04018%2C2509.04908%2C2509.05368%2C2509.05831%2C2509.06917%2C2509.07098%2C2509.07553%2C2509.08820%2C2509.10345%2C2509.10540%2C2509.11253%2C2509.12543%2C2509.12987%2C2509.13615%2C2509.14547%2C2509.14744%2C2509.15061%2C2509.15237%2C2509.17336%2C2509.19185%2C2509.19199%2C2509.21072%2C2509.21291%2C2509.21766%2C2509.21799%2C2509.22715%2C2509.22830%2C2509.23738%2C2509.24704%2C2509.25455%2C2509.25693%2C2510.00172%2C2510.00515%2C2510.02190%2C2510.04491%2C2510.05096%2C2510.05156%2C2510.05788%2C2510.07284%2C2510.07915%2C2510.08787%2C2510.09517&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T11:20:50Z", "role": "primary", "sha256": "776192081826494b05077d71757f3006575ed50a8cc244004e2de7053104d7eb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://arxiv.org/abs/2509.07553v3"}], "id": "arxiv:2509.07553", "identifiers": {"arxiv": "2509.07553"}, "kind": "paper", "last_verified_at": "2026-10-17T11:20:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2509.07553"}], "name": "VeriOS: Query-Driven Proactive Human-Agent-GUI Interaction for Trustworthy OS Agents", "notes": "Reviewed against the exact-version official arXiv abstract page; its complete title, ordered author list, visible version, and normalized abstract SHA match the retained snapshot. The original API response body was not locally available. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-09-09", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces VeriOS-Agent, which learns when to execute GUI actions autonomously and when to query a user under untrustworthy conditions. Its staged training combines supervised learning and group-relative policy optimization.", "topics": ["human-in-the-loop", "safety-security", "reinforcement-learning"], "updated_at": "2026-04-03", "url": "https://arxiv.org/abs/2509.07553"}
{"aliases": [], "authors": ["Haitao Hu", "Peng Chen", "Yanpeng Zhao", "Yuqi Chen"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "ae582aceb31075c460a9e4a6dce9978574c3e0ada731b26c13738a4a4fafcc32", "submitted_at": "2025-09-09T13:59:00Z", "updated_at": "2025-09-09T13:59:00Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2509.07764v1"}}, "environments": ["desktop", "linux", "web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "topics", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T11:25:42Z", "role": "primary", "sha256": "7f303629ef6d9c7df128c6d12d472abd9bf238e0204515f18839249adef317b9", "supports": ["summary", "environments", "scope", "notes"], "url": "https://arxiv.org/html/2509.07764v1"}], "id": "arxiv:2509.07764", "identifiers": {"arxiv": "2509.07764"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2509.07764"}], "name": "AgentSentinel: An End-to-End and Real-Time Security Defense Framework for Computer-Use Agents", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced. The official full paper confirms GUI/desktop interaction alongside Bash and file tools. Its abstract and body enumerate attack categories differently; no category count is promoted into this annotation.", "published_at": "2025-09-09", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces AgentSentinel, which intercepts sensitive computer-agent operations and audits them using task context together with system traces. BadComputerUse evaluates this enforcement approach across attack scenarios involving tool and computer control.", "topics": ["safety-security", "observability", "evaluation"], "updated_at": "2025-09-09", "url": "https://arxiv.org/abs/2509.07764"}
{"aliases": [], "authors": ["Jingyu Tang", "Chaoran Chen", "Jiawen Li", "Zhiping Zhang", "Bingcan Guo", "Ibrahim Khalilov", "Simret Araya Gebreegziabher", "Bingsheng Yao", "Dakuo Wang", "Yanfang Ye", "Tianshi Li", "Ziang Xiao", "Yaxing Yao", "Toby Jia-Jun Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "247ed053a4c8f62106d9f8e9af0801d94265ed712f4ae9f095fd8aaccbe33c59", "submitted_at": "2025-09-12T22:26:31Z", "updated_at": "2025-09-12T22:26:31Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2509.10723v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2509.10723", "identifiers": {"arxiv": "2509.10723"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2509.10723"}], "name": "Dark Patterns Meet GUI Agents: LLM Agent Susceptibility to Manipulative Interfaces and the Role of Human Oversight", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-09-12", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies how GUI agents, people, and human-agent teams respond to manipulative interface designs. The two-phase evaluation examines failure modes and the benefits and costs of human oversight.", "topics": ["human-in-the-loop", "safety-security", "evaluation"], "updated_at": "2025-09-12", "url": "https://arxiv.org/abs/2509.10723"}
{"aliases": [], "authors": ["Yitong Zhang", "Ximo Li", "Liyi Cai", "Jia Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "9ab96edd8e52b3a490455b46878c92c23d15c0f9e105e71dbd6848e058da42ec", "submitted_at": "2025-09-14T12:47:54Z", "updated_at": "2026-08-24T12:04:58Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2509.11250v3"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2509.11250", "identifiers": {"arxiv": "2509.11250"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2509.11250"}], "name": "Environmental Injection Attacks against GUI Agents in Realistic Dynamic Environments", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-09-14", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces Chameleon, an environmental-injection evaluation framework that varies webpage content and trigger context. It combines simulated page variation with attention-based optimization and tests GUI agents across realistic websites.", "topics": ["safety-security", "robustness", "evaluation"], "updated_at": "2026-08-24", "url": "https://arxiv.org/abs/2509.11250"}
{"aliases": [], "authors": ["Shuning Zhang", "Yutong Jiang", "Rongjun Ma", "Yuting Yang", "Mingyao Xu", "Zhixin Huang", "Xin Yi", "Hewu Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "f37ebc10535f96d17d28990e608a1f2eb6cc004f40d58b3288c4a81d781160ea", "submitted_at": "2025-09-15T13:58:52Z", "updated_at": "2025-09-15T13:58:52Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2509.11939v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2509.11939", "identifiers": {"arxiv": "2509.11939"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2509.11939"}], "name": "PrivWeb: Unobtrusive and Content-aware Privacy Protection For Web Agents", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-09-15", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces PrivWeb, which uses a local language model to anonymize interface information according to user preferences. Adaptive notifications and selective pauses give users control over sensitive information access during web-agent tasks.", "topics": ["safety-security", "human-in-the-loop", "web-navigation"], "updated_at": "2025-09-15", "url": "https://arxiv.org/abs/2509.11939"}
{"aliases": [], "authors": ["Liangtao Lin", "Zhaomeng Zhu", "Tianwei Zhang", "Yonggang Wen"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "2c1a4019df77278943829e16b155771abb1c8ca30726044cf5969227e0437f5f", "submitted_at": "2025-09-17T05:14:11Z", "updated_at": "2025-09-17T05:14:11Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2509.13704v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2509.13704", "identifiers": {"arxiv": "2509.13704"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2509.13704"}], "name": "InfraMind: A Novel Exploration-based GUI Agentic Framework for Mission-critical Industrial Management", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-09-17", "relations": [], "review_status": "curated", "scope": "core", "summary": "InfraMind explores industrial management GUIs using virtual-machine snapshots and builds memory for planning and hierarchical state recognition. Its framework also studies knowledge distillation and safety checks on open-source and commercial data-center management platforms.", "topics": ["industry", "planning", "memory", "safety-security"], "updated_at": "2025-09-17", "url": "https://arxiv.org/abs/2509.13704"}
{"aliases": [], "authors": ["Junliang Liu", "Jingyu Xiao", "Wenxin Tang", "Zhixian Wang", "Zipeng Xie", "Wenxuan Wang", "Minrui Zhang", "Shuanghe Yu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "8fed6f081e0330c54a304c79e8a4183865aae48f0e944545029d3de9852f8028", "submitted_at": "2025-09-26T02:38:14Z", "updated_at": "2026-03-04T16:31:39Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2509.21782v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2509.21782", "identifiers": {"arxiv": "2509.21782"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2509.21782"}], "name": "Benchmarking MLLM-based Web Understanding: Reasoning, Robustness and Safety", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-09-26", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates multimodal LLMs on web interface reasoning, visual perturbation robustness, and safety-critical web understanding tasks.", "topics": ["evaluation", "screen-understanding", "safety-security", "robustness"], "updated_at": "2026-03-04", "url": "https://arxiv.org/abs/2509.21782"}
{"aliases": [], "authors": ["Haochen Gong", "Chenxiao Li", "Rui Chang", "Wenbo Shen"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "922c1d9b60db7c17b9019a320a4cfc2dd4eac325fd946f024ca82d43785f5eba", "submitted_at": "2025-09-26T12:19:27Z", "updated_at": "2026-01-14T06:07:30Z", "version": 4, "versioned_url": "https://arxiv.org/abs/2509.22256v4"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:50Z", "role": "primary", "sha256": "26fb4922c4c0bb5c9069935ed0aef0a9f56938cf8ae1af9724894fa1409955fb", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=600&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2509.22256", "identifiers": {"arxiv": "2509.22256"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:50Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2509.22256"}], "name": "Secure and Efficient Access Control for Computer-Use Agents via Context Space", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2025-09-26", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces CSAgent, which combines static access policies with user intent and context and enforces them through an operating-system service. A supporting toolchain constructs and refines policies for agents using API, CLI and GUI interfaces.", "topics": ["safety-security", "tool-integration"], "updated_at": "2026-01-14", "url": "https://arxiv.org/abs/2509.22256"}
{"aliases": [], "authors": ["Yinuo Liu", "Xilong Wang", "Ruohan Xu", "Yuqi Jia", "Neil Zhenqiang Gong"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e997e87268cfe66869bf19fdc8445e5487c980736a63cbe76f779bd1f142c396", "submitted_at": "2025-10-01T18:34:06Z", "updated_at": "2026-09-23T16:44:27Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2510.01354v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2510.01354", "identifiers": {"arxiv": "2510.01354"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2510.01354"}], "name": "WAInjectBench: Benchmarking Prompt Injection Detections for Web Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced. The reported contribution evaluates attack detection; end-to-end prevention of browser-agent compromise is not inferred.", "published_at": "2025-10-01", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces WAInjectBench with a threat-model taxonomy and malicious and benign text-image datasets for web-agent prompt-injection detection. It compares detector families across explicit and less visible forms of injected content.", "topics": ["safety-security", "evaluation", "robustness"], "updated_at": "2026-09-23", "url": "https://arxiv.org/abs/2510.01354"}
{"aliases": [], "authors": ["Erfan Shayegani", "Keegan Hines", "Yue Dong", "Nael Abu-Ghazaleh", "Roman Lutz", "Spencer Whitehead", "Vidhisha Balachandran", "Besmira Nushi", "Vibhav Vineet"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "a6426fa5eeb3dfafef30e47dd46a6a43dff3e24bf0ff485b7d10ab9efe133f94", "submitted_at": "2025-10-02T04:52:15Z", "updated_at": "2025-10-02T04:52:15Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2510.01670v1"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2510.01670", "identifiers": {"arxiv": "2510.01670"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2510.01670"}], "name": "Just Do It!? Computer-Use Agents Exhibit Blind Goal-Directedness", "notes": "The annotation was reviewed against the complete retained primary abstract. Experiments and released artifacts have not been independently reproduced.", "published_at": "2025-10-02", "relations": [], "review_status": "curated", "scope": "core", "summary": "BLIND-ACT tests whether computer-use agents persist with ambiguous, contradictory, or infeasible goals without adequate contextual reasoning. Its OSWorld-based tasks and judged trajectories examine these behaviors and the effect of mitigation prompts.", "topics": ["evaluation", "safety-security", "robustness"], "updated_at": "2025-10-02", "url": "https://arxiv.org/abs/2510.01670"}
{"aliases": [], "authors": ["Renhua Ding", "Xiao Yang", "Zhengwei Fang", "Jun Luo", "Kun He", "Jun Zhu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e80999148931b7485d51ef87887d06240f813747a9739282ed0daf53a912f9f6", "submitted_at": "2025-10-09T05:34:57Z", "updated_at": "2026-04-09T01:41:48Z", "version": 4, "versioned_url": "https://arxiv.org/abs/2510.07809v4"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2510.07809", "identifiers": {"arxiv": "2510.07809"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2510.07809"}], "name": "Invisible to Humans, Triggered by Agents: Stealthy Jailbreak Attacks on Mobile Vision-Language Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-10-09", "relations": [], "review_status": "curated", "scope": "core", "summary": "Uses the near-zero contact-touch signals of automated mobile agents to expose malicious visual content selectively during agent interactions. The paper combines this agent-only perceptual injection with HG-IDA* prompt optimization to study unauthorized cross-app actions.", "topics": ["safety-security", "mobile-navigation", "robustness"], "updated_at": "2026-04-09", "url": "https://arxiv.org/abs/2510.07809"}
{"aliases": [], "authors": ["Zonghao Ying", "Yangguang Shao", "Jianle Gan", "Gan Xu", "Wenxin Zhang", "Quanchen Zou", "Junzheng Shi", "Zhenfei Yin", "Mingchuan Zhang", "Aishan Liu", "Xianglong Liu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "43c9f7c91539df587a4f08888820731aef5bb3fca2cfb9e742443330aaf1c35f", "submitted_at": "2025-10-11T07:18:12Z", "updated_at": "2026-04-14T02:53:37Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2510.10073v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2510.10073", "identifiers": {"arxiv": "2510.10073"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2510.10073"}], "name": "SecureWebArena: A Holistic Security Evaluation Benchmark for LVLM-based Web Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2025-10-11", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces SecureWebArena with simulated interactive websites, an attack-vector taxonomy and labeled agent trajectories. Its evaluation separates failures in reasoning, behavioral traces and task outcomes across vision-language agent families.", "topics": ["safety-security", "evaluation", "robustness"], "updated_at": "2026-04-14", "url": "https://arxiv.org/abs/2510.10073"}
{"aliases": [], "authors": ["Longjie Guo", "Chenjie Yuan", "Mingyuan Zhong", "Robert Wolfe", "Ruican Zhong", "Yue Xu", "Bingbing Wen", "Hua Shen", "Lucy Lu Wang", "Alexis Hiniker"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e6152a61f81c435e76a432bc10086827c0cf0d2f44026bac03d6b80ebae8ca77", "submitted_at": "2025-10-13T06:12:03Z", "updated_at": "2026-02-23T08:31:13Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2510.11035v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2510.11035", "identifiers": {"arxiv": "2510.11035"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2510.11035"}], "name": "SusBench: An Online Benchmark for Evaluating Dark Pattern Susceptibility of Computer-Use Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2025-10-13", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces SusBench, an online benchmark that inserts dark-pattern designs into consumer websites for agent evaluation. A human study checks the realism of the interventions, and agent trials compare susceptibility across pattern types.", "topics": ["ui-design", "safety-security", "evaluation"], "updated_at": "2026-02-23", "url": "https://arxiv.org/abs/2510.11035"}
{"aliases": [], "authors": ["Xiaoxue Ren", "Penghao Jiang", "Kaixin Li", "Zhiyong Huang", "Xiaoning Du", "Jiaojiao Jiang", "Zhenchang Xing", "Jiamou Sun", "Terry Yue Zhuo"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e92db1e568c3c512aa00e2363bbdc9869cf834d638c331b45d4753d29a282004", "submitted_at": "2025-10-14T06:52:15Z", "updated_at": "2025-10-14T06:52:15Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2510.12200v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2510.12200", "identifiers": {"arxiv": "2510.12200"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2510.12200"}], "name": "HackWorld: Evaluating Computer-Use Agents on Exploiting Web Application Vulnerabilities", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2025-10-14", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces HackWorld for evaluating visual computer-use agents in a capture-the-flag collection of web applications. The benchmark examines interface navigation and multistep planning during vulnerability discovery and exploitation tasks.", "topics": ["safety-security", "evaluation", "web-navigation"], "updated_at": "2025-10-14", "url": "https://arxiv.org/abs/2510.12200"}
{"aliases": [], "authors": ["Avihay Cohen"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "bedde658a28c57551c27563fa22cb037839dddd4e5eabe1c66afa27523d0d406", "submitted_at": "2025-10-15T13:39:13Z", "updated_at": "2025-10-15T13:39:13Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2510.13543v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:32:52Z", "role": "primary", "sha256": "0b730323b2ce85c39aa7d534665901a480ff06540c73f1f80832fcc0735b8789", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2310.08588%2C2402.09205%2C2406.05720%2C2407.21075%2C2410.07166%2C2503.03743%2C2508.04037%2C2508.10833%2C2508.11360%2C2508.13024%2C2508.13186%2C2508.14104%2C2509.01619%2C2509.01720%2C2509.02322%2C2509.04198%2C2509.07764%2C2509.21823%2C2509.22256%2C2509.23263%2C2509.23866%2C2509.25047%2C2509.26539%2C2510.00413%2C2510.00536%2C2510.01524%2C2510.02204%2C2510.02250%2C2510.02418%2C2510.03204%2C2510.03612%2C2510.04039%2C2510.04373%2C2510.04607%2C2510.06223%2C2510.06607%2C2510.09872%2C2510.10666%2C2510.11035%2C2510.11221%2C2510.11892%2C2510.13543%2C2510.15455%2C2510.15863%2C2510.16252%2C2510.18113%2C2510.18314%2C2510.19205%2C2510.22732%2C2511.06101&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:32:52Z", "role": "primary", "sha256": "0b730323b2ce85c39aa7d534665901a480ff06540c73f1f80832fcc0735b8789", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2310.08588%2C2402.09205%2C2406.05720%2C2407.21075%2C2410.07166%2C2503.03743%2C2508.04037%2C2508.10833%2C2508.11360%2C2508.13024%2C2508.13186%2C2508.14104%2C2509.01619%2C2509.01720%2C2509.02322%2C2509.04198%2C2509.07764%2C2509.21823%2C2509.22256%2C2509.23263%2C2509.23866%2C2509.25047%2C2509.26539%2C2510.00413%2C2510.00536%2C2510.01524%2C2510.02204%2C2510.02250%2C2510.02418%2C2510.03204%2C2510.03612%2C2510.04039%2C2510.04373%2C2510.04607%2C2510.06223%2C2510.06607%2C2510.09872%2C2510.10666%2C2510.11035%2C2510.11221%2C2510.11892%2C2510.13543%2C2510.15455%2C2510.15863%2C2510.16252%2C2510.18113%2C2510.18314%2C2510.19205%2C2510.22732%2C2511.06101&max_results=50"}], "id": "arxiv:2510.13543", "identifiers": {"arxiv": "2510.13543"}, "kind": "paper", "last_verified_at": "2026-10-16T06:32:52Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2510.13543"}], "name": "In-Browser LLM-Guided Fuzzing for Real-Time Prompt Injection Testing in Agentic AI Browsers", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced. The short abstract supplies the framework and threat setting but no quantitative evaluation details.", "published_at": "2025-10-15", "relations": [], "review_status": "curated", "scope": "core", "summary": "Proposes an in-browser fuzzing framework that uses language-model guidance to test indirect prompt-injection vulnerabilities in agentic browsers. The work focuses on unwanted actions triggered by webpage instructions under the browser agent's user privileges.", "topics": ["safety-security", "automation"], "updated_at": "2025-10-15", "url": "https://arxiv.org/abs/2510.13543"}
{"aliases": [], "authors": ["Gucongcong Fan", "Chaoyue Niu", "Chengfei Lyu", "Fan Wu", "Guihai Chen"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "8584797a1c12158b20da04c30e3a0f577c63f02f9e906a051579959de2cae480", "submitted_at": "2025-10-17T09:11:05Z", "updated_at": "2025-10-17T09:11:05Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2510.15455v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:32:52Z", "role": "primary", "sha256": "0b730323b2ce85c39aa7d534665901a480ff06540c73f1f80832fcc0735b8789", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2310.08588%2C2402.09205%2C2406.05720%2C2407.21075%2C2410.07166%2C2503.03743%2C2508.04037%2C2508.10833%2C2508.11360%2C2508.13024%2C2508.13186%2C2508.14104%2C2509.01619%2C2509.01720%2C2509.02322%2C2509.04198%2C2509.07764%2C2509.21823%2C2509.22256%2C2509.23263%2C2509.23866%2C2509.25047%2C2509.26539%2C2510.00413%2C2510.00536%2C2510.01524%2C2510.02204%2C2510.02250%2C2510.02418%2C2510.03204%2C2510.03612%2C2510.04039%2C2510.04373%2C2510.04607%2C2510.06223%2C2510.06607%2C2510.09872%2C2510.10666%2C2510.11035%2C2510.11221%2C2510.11892%2C2510.13543%2C2510.15455%2C2510.15863%2C2510.16252%2C2510.18113%2C2510.18314%2C2510.19205%2C2510.22732%2C2511.06101&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:32:52Z", "role": "primary", "sha256": "0b730323b2ce85c39aa7d534665901a480ff06540c73f1f80832fcc0735b8789", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2310.08588%2C2402.09205%2C2406.05720%2C2407.21075%2C2410.07166%2C2503.03743%2C2508.04037%2C2508.10833%2C2508.11360%2C2508.13024%2C2508.13186%2C2508.14104%2C2509.01619%2C2509.01720%2C2509.02322%2C2509.04198%2C2509.07764%2C2509.21823%2C2509.22256%2C2509.23263%2C2509.23866%2C2509.25047%2C2509.26539%2C2510.00413%2C2510.00536%2C2510.01524%2C2510.02204%2C2510.02250%2C2510.02418%2C2510.03204%2C2510.03612%2C2510.04039%2C2510.04373%2C2510.04607%2C2510.06223%2C2510.06607%2C2510.09872%2C2510.10666%2C2510.11035%2C2510.11221%2C2510.11892%2C2510.13543%2C2510.15455%2C2510.15863%2C2510.16252%2C2510.18113%2C2510.18314%2C2510.19205%2C2510.22732%2C2511.06101&max_results=50"}], "id": "arxiv:2510.15455", "identifiers": {"arxiv": "2510.15455"}, "kind": "paper", "last_verified_at": "2026-10-16T06:32:52Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2510.15455"}], "name": "CORE: Reducing UI Exposure in Mobile Agents via Collaboration Between Cloud and Local LLMs", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced. XML screen hierarchies do not by themselves establish Android deployment.", "published_at": "2025-10-17", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces CORE, which divides mobile UI hierarchies into semantic blocks and shares planning between local and cloud language models. Local ranking limits the screen information sent for cloud-side element selection, with a multi-round mechanism for correcting inadequate context.", "topics": ["safety-security", "mobile-navigation", "efficiency"], "updated_at": "2025-10-17", "url": "https://arxiv.org/abs/2510.15455"}
{"aliases": [], "authors": ["Devin Ersoy", "Brandon Lee", "Ananth Shreekumar", "Arjun Arunasalam", "Muhammad Ibrahim", "Antonio Bianchi", "Z. Berkay Celik"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "db3c25ba55221ca3d7c45ce4243a4a37b56a698909c0ea3eaaabdcd37388a1ec", "submitted_at": "2025-10-20T21:26:26Z", "updated_at": "2025-10-20T21:26:26Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2510.18113v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2510.18113", "identifiers": {"arxiv": "2510.18113"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2510.18113"}], "name": "Investigating the Impact of Dark Patterns on LLM-Based Web Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2025-10-20", "relations": [], "review_status": "curated", "scope": "core", "summary": "Presents LiteAgent for recording web-agent interactions and TrickyArena for enabling or disabling dark patterns in controlled websites. Experiments compare how individual patterns, combined patterns and visual or HTML changes affect agent decisions.", "topics": ["safety-security", "ui-design", "evaluation"], "updated_at": "2025-10-20", "url": "https://arxiv.org/abs/2510.18113"}
{"aliases": [], "authors": ["Qiushi Sun", "Mukai Li", "Zhoumianze Liu", "Zhihui Xie", "Fangzhi Xu", "Zhangyue Yin", "Kanzhi Cheng", "Zehao Li", "Zichen Ding", "Qi Liu", "Zhiyong Wu", "Zhuosheng Zhang", "Ben Kao", "Lingpeng Kong"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "7e3fde732c9e760437f2c60d585f7ba0d2ed13bbe9e051838844a21174159099", "submitted_at": "2025-10-28T13:22:39Z", "updated_at": "2026-07-26T01:16:06Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2510.24411v3"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2510.24411", "identifiers": {"arxiv": "2510.24411"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2510.24411"}], "name": "OS-Sentinel: Towards Safety-Enhanced Mobile GUI Agents via Hybrid Validation in Realistic Workflows", "notes": "Reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2025-10-28", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces MobileRisk-Live and OS-Sentinel, combining a formal verifier for system-level violations with a multimodal judge for contextual risks in mobile GUI workflows.", "topics": ["safety-security", "evaluation", "mobile-navigation"], "updated_at": "2026-07-26", "url": "https://arxiv.org/abs/2510.24411"}
{"aliases": [], "authors": ["Chenghao Du", "Quanfeng Huang", "Tingxuan Tang", "Zihao Wang", "Adwait Nadkarni", "Yue Xiao"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "61b611712fff54af25aae660f2353fa6c3aaa764b0767831ebf2278c120d9dea", "submitted_at": "2025-10-31T03:35:59Z", "updated_at": "2026-07-07T02:18:26Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2510.27140v3"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2510.27140", "identifiers": {"arxiv": "2510.27140"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2510.27140"}], "name": "Measuring the Security of Mobile LLM Agents under Adversarial Prompts from Untrusted Third-Party Channels", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-10-31", "relations": [], "review_status": "curated", "scope": "core", "summary": "Measures how untrusted third-party notifications, messages, and app content can hijack mobile LLM agents into executing unauthorized cross-app actions.", "topics": ["safety-security", "mobile-navigation", "evaluation"], "updated_at": "2026-07-07", "url": "https://arxiv.org/abs/2510.27140"}
{"aliases": [], "authors": ["Fuyao Zhang", "Jiaming Zhang", "Che Wang", "Xiongtao Sun", "Yurong Hao", "Guowei Guan", "Wenjie Li", "Longtao Huang", "Wei Yang Bryan Lim"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "d918a5754fe907fc273639f3277a3566d649c08da244714d0d7737456bede230", "submitted_at": "2025-11-17T11:08:45Z", "updated_at": "2025-11-17T11:08:45Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2511.13248v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2511.13248", "identifiers": {"arxiv": "2511.13248"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2511.13248"}], "name": "DualTAP: A Dual-Task Adversarial Protector for Mobile MLLM Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2025-11-17", "relations": [], "review_status": "curated", "scope": "core", "summary": "Applies dual-task adversarial perturbations to mobile screenshots to conceal personally identifiable information from untrusted MLLM routers while preserving agent task utility.", "topics": ["safety-security", "mobile-navigation"], "updated_at": "2025-11-17", "url": "https://arxiv.org/abs/2511.13248"}
{"aliases": [], "authors": ["Aram Vardanyan"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "17debfa3b59e5a2fcae950c2659e5851ae101cc58181e0cc4e0320befe9b4cb1", "submitted_at": "2025-11-22T12:18:35Z", "updated_at": "2025-11-22T12:18:35Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2511.19477v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2511.19477", "identifiers": {"arxiv": "2511.19477"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2511.19477"}], "name": "Building Browser Agents: Architecture, Security, and Practical Solutions", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced. Broader claims about the safety of general-purpose browsing are the paper's position, not a catalog conclusion.", "published_at": "2025-11-22", "relations": [], "review_status": "curated", "scope": "core", "summary": "Reports architectural and security lessons from building a browser agent, including hybrid accessibility-tree and visual context, browser tooling and programmatic constraints. A WebGames evaluation accompanies the discussion of reliability and specialization.", "topics": ["web-navigation", "safety-security", "tool-integration"], "updated_at": "2025-11-22", "url": "https://arxiv.org/abs/2511.19477"}
{"aliases": [], "authors": ["Alisha Ukani", "Hamed Haddadi", "Ali Shahin Shamsabadi", "Peter Snyder"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "2576701bbdc4fd8f1f571ed691b06185d1e16bf4955a5c99b10200d796ba6525", "submitted_at": "2025-12-08T17:16:12Z", "updated_at": "2025-12-08T17:16:12Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2512.07725v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2512.07725", "identifiers": {"arxiv": "2512.07725"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2512.07725"}], "name": "Privacy Practices of Browser Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced. The abstract describes artifact release as planned; the annotation does not independently verify availability.", "published_at": "2025-12-08", "relations": [], "review_status": "curated", "scope": "core", "summary": "Presents a privacy evaluation framework for browser agents covering component weaknesses, website defenses, tracking, privacy-related prompts and personal-data leakage. The framework is applied to deployed browser-agent tools.", "topics": ["safety-security", "evaluation", "web-navigation"], "updated_at": "2025-12-08", "url": "https://arxiv.org/abs/2512.07725"}
{"aliases": [], "authors": ["Luoxi Meng", "Henry Feng", "Ilia Shumailov", "Earlence Fernandes"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "fa6b4d9fa34610ec6fa8c0ac8b075127365aad57a20d814a2e838a223cf8bd56", "submitted_at": "2025-12-14T08:25:31Z", "updated_at": "2026-03-04T01:49:57Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2512.12594v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-17T12:31:09Z", "role": "primary", "sha256": "f8cee8b281f54ecb3112bb26b59d166617d33f08f098f44818968a9c9b777ce6", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2512.06866%2C2512.07287%2C2512.08769%2C2512.12594%2C2512.12730%2C2512.13159%2C2512.14012%2C2512.16925%2C2512.16962%2C2512.16975%2C2512.21373%2C2512.23647%2C2512.23738%2C2601.01743%2C2601.02046%2C2601.02163%2C2601.02454%2C2601.02702%2C2601.03468%2C2601.03928%2C2601.04795%2C2601.05107%2C2601.05432%2C2601.06411%2C2601.06487%2C2601.06490%2C2601.06565%2C2601.06663%2C2601.07143%2C2601.07375%2C2601.08012%2C2601.08079%2C2601.08225%2C2601.08237%2C2601.08731%2C2601.09113%2C2601.09749%2C2601.09883%2C2601.10378%2C2601.10440%2C2601.11653%2C2601.11895%2C2601.11903%2C2601.11974%2C2601.12042%2C2601.12307%2C2601.12560%2C2601.12641%2C2601.13186%2C2601.13383&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T12:31:09Z", "role": "primary", "sha256": "f8cee8b281f54ecb3112bb26b59d166617d33f08f098f44818968a9c9b777ce6", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?id_list=2512.06866%2C2512.07287%2C2512.08769%2C2512.12594%2C2512.12730%2C2512.13159%2C2512.14012%2C2512.16925%2C2512.16962%2C2512.16975%2C2512.21373%2C2512.23647%2C2512.23738%2C2601.01743%2C2601.02046%2C2601.02163%2C2601.02454%2C2601.02702%2C2601.03468%2C2601.03928%2C2601.04795%2C2601.05107%2C2601.05432%2C2601.06411%2C2601.06487%2C2601.06490%2C2601.06565%2C2601.06663%2C2601.07143%2C2601.07375%2C2601.08012%2C2601.08079%2C2601.08225%2C2601.08237%2C2601.08731%2C2601.09113%2C2601.09749%2C2601.09883%2C2601.10378%2C2601.10440%2C2601.11653%2C2601.11895%2C2601.11903%2C2601.11974%2C2601.12042%2C2601.12307%2C2601.12560%2C2601.12641%2C2601.13186%2C2601.13383&max_results=50"}], "id": "arxiv:2512.12594", "identifiers": {"arxiv": "2512.12594"}, "kind": "paper", "last_verified_at": "2026-10-17T12:31:09Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2512.12594"}], "name": "ceLLMate: Sandboxing Browser AI Agents", "notes": "Reviewed against the complete retained official abstract. Experimental findings and release availability have not been independently reproduced or verified.", "published_at": "2025-12-14", "relations": [], "review_status": "curated", "scope": "core", "summary": "ceLLMate constrains browser-agent authority through an extension that applies policies at the HTTP layer. It evaluates this sandbox against prompt-injection attacks in WASP, avoiding dependence on brittle low-level click and keystroke policies.", "topics": ["safety-security", "sandbox", "web-navigation", "robustness", "evaluation"], "updated_at": "2026-03-04", "url": "https://arxiv.org/abs/2512.12594"}
{"aliases": [], "authors": ["Karolina Korgul", "Yushi Yang", "Arkadiusz Drohomirecki", "Piotr Błaszczyk", "Will Howard", "Lukas Aichberger", "Chris Russell", "Philip H. S. Torr", "Adam Mahdi", "Adel Bibi"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "a255cccbe3f16f498918126724c391911e1d745c093189790cbab715fd28a2d2", "submitted_at": "2025-12-29T01:09:10Z", "updated_at": "2026-07-02T16:12:04Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2512.23128v3"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:47Z", "role": "primary", "sha256": "ae98e77b8bb125e45277aacec768a047e7bb0e13f33ca9ff6ca7c19e29e56cca", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=500&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2512.23128", "identifiers": {"arxiv": "2512.23128"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:47Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2512.23128"}], "name": "It's a TRAP! Task-Redirecting Agent Persuasion Benchmark for Web Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2025-12-29", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces TRAP, which evaluates task-redirection attacks embedded in realistic web interfaces. Controlled website clones and a modular injection framework support comparisons of persuasion techniques and contextual changes across agents.", "topics": ["safety-security", "evaluation", "ui-design"], "updated_at": "2026-07-02", "url": "https://arxiv.org/abs/2512.23128"}
{"aliases": [], "authors": ["Xinyi Wu", "Geng Hong", "Yueyue Chen", "MingXuan Liu", "Feier Jin", "Xudong Pan", "Jiarun Dai", "Baojun Liu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "4af00e0632ee21f70409998fc22f67e9a4ed1de415b488eda122c6ef94cab7aa", "submitted_at": "2026-01-12T07:10:08Z", "updated_at": "2026-01-12T07:10:08Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2601.07263v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2601.07263", "identifiers": {"arxiv": "2601.07263"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2601.07263"}], "name": "When Bots Take the Bait: Exposing and Mitigating the Emerging Social Engineering Attack in Web Automation Agent", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-01-12", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies deceptive social-engineering web patterns targeting browser automation agents and proposes defenses against bait-based manipulation.", "topics": ["safety-security", "web-navigation", "robustness"], "updated_at": "2026-01-12", "url": "https://arxiv.org/abs/2601.07263"}
{"aliases": [], "authors": ["Xinyi Wu", "Jiagui Chen", "Geng Hong", "Jiayi Dong", "Xudong Pan", "Jiarun Dai", "Min Yang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "626702d22786336ff65b6e89d97e2034ff7aa144fab4ecc286d169fca6993410", "submitted_at": "2026-01-13T10:21:28Z", "updated_at": "2026-01-13T10:21:28Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2601.08406v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2601.08406", "identifiers": {"arxiv": "2601.08406"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:43Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2601.08406"}], "name": "WebTrap Park: An Automated Platform for Systematic Security Evaluation of Web Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-01-13", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces WebTrap Park, a platform that turns web-agent security risks into executable tasks and assesses concrete interactions with live pages. It compares agent frameworks without requiring modifications to their implementations.", "topics": ["safety-security", "evaluation", "web-navigation"], "updated_at": "2026-01-13", "url": "https://arxiv.org/abs/2601.08406"}
{"aliases": [], "authors": ["Hanna Foerster", "Tom Blanchard", "Kristina Nikolić", "Ilia Shumailov", "Cheng Zhang", "Robert Mullins", "Nicolas Papernot", "Florian Tramèr", "Yiren Zhao"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e32917bc92b65701eea06c09a2c681c7504418f8ce888af2b72012a1da50c9f6", "submitted_at": "2026-01-14T23:06:35Z", "updated_at": "2026-06-04T15:38:02Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2601.09923v3"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2601.09923", "identifiers": {"arxiv": "2601.09923"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:43Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2601.09923"}], "name": "CaMeLs Can Use Computers Too: System-level Security for Computer Use Agents", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and released artifacts have not been independently reproduced.", "published_at": "2026-01-14", "relations": [], "review_status": "curated", "scope": "core", "summary": "NOVA separates trusted task planning from screen-derived observations by producing a branching plan before execution and resolving runtime values through perception. Its OSWorld study examines security and utility while identifying remaining risks from perception-driven branch steering.", "topics": ["safety-security", "planning", "robustness"], "updated_at": "2026-06-04", "url": "https://arxiv.org/abs/2601.09923"}
{"aliases": [], "authors": ["Yi Qian", "Kunwei Qian", "Xingbang He", "Ligeng Chen", "Jikang Zhang", "Tiantai Zhang", "Haiyang Wei", "Linzhang Wang", "Hao Wu", "Bing Mao"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "21848dea9e000125550fff7fe3844a57bf100215bf9eb800cbdbb612675f7074", "submitted_at": "2026-01-18T10:54:54Z", "updated_at": "2026-07-15T15:32:11Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2601.12349v3"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2601.12349", "identifiers": {"arxiv": "2601.12349"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:43Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2601.12349"}], "name": "Mind the Gap: Action Rebinding Attacks against Android GUI Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-01-18", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies action-rebinding attacks in which an Android GUI agent’s planned input is redirected to a different application during the observation–action delay. The evaluation examines cross-application execution and the consequences of treating an earlier screen as the current action target.", "topics": ["safety-security", "cross-app", "evaluation"], "updated_at": "2026-07-15", "url": "https://arxiv.org/abs/2601.12349"}
{"aliases": [], "authors": ["Wenqi Zhang", "Yulin Shen", "Changyue Jiang", "Jiarun Dai", "Geng Hong", "Xudong Pan"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "b57586682d4675387dea2687e7c15d70cc10aecac883e35fe1812c39abcbc1dc", "submitted_at": "2026-01-19T08:32:09Z", "updated_at": "2026-01-19T08:32:09Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2601.12822v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2601.12822", "identifiers": {"arxiv": "2601.12822"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:43Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2601.12822"}], "name": "MirrorGuard: Toward Secure Computer-Use Agents via Simulation-to-Real Reasoning Correction", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and released artifacts have not been independently reproduced.", "published_at": "2026-01-19", "relations": [], "review_status": "curated", "scope": "core", "summary": "MirrorGuard trains a reasoning-correction defense using GUI risk trajectories generated in a text-based neural-symbolic simulator. It evaluates whether corrections transfer to computer-use agents operating real interfaces, keeping simulated training separate from deployment tests.", "topics": ["safety-security", "data-generation", "robustness"], "updated_at": "2026-01-19", "url": "https://arxiv.org/abs/2601.12822"}
{"aliases": [], "authors": ["Yanxi Wang", "Zhiling Zhang", "Wenbo Zhou", "Weiming Zhang", "Jie Zhang", "Qiannan Zhu", "Yu Shi", "Shuxin Zheng", "Jiyan He"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "1ca912ad67a681e8e908a5a403e982f39b5080bee4df376eb0b598468f3bb504", "submitted_at": "2026-01-26T11:33:40Z", "updated_at": "2026-05-13T07:11:36Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2601.18842v3"}}, "environments": ["android", "desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2601.18842", "identifiers": {"arxiv": "2601.18842"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:43Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2601.18842"}], "name": "GUIGuard-Bench: Toward a General Evaluation for Privacy-Preserving GUI Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-01-26", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces GUIGuard-Bench, which annotates private regions in GUI trajectories with categories, risk levels, and relevance to the user’s task. It evaluates privacy recognition and the effect of screenshot protection on offline planning and task utility.", "topics": ["safety-security", "evaluation", "grounding"], "updated_at": "2026-05-13", "url": "https://arxiv.org/abs/2601.18842"}
{"aliases": [], "authors": ["Yurun Chen", "Zeyi Liao", "Ping Yin", "Taotao Xie", "Keting Yin", "Shengyu Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "d1a6f98c0ab0a221c169796d9680e4b2a7bf7a0775c0493254f47d5e0a9b3c4e", "submitted_at": "2026-02-02T07:04:06Z", "updated_at": "2026-02-02T07:04:06Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2602.01725v1"}}, "environments": ["desktop", "web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "kind", "summary", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T11:22:38+00:00", "role": "primary", "sha256": "6e1266c5a47cd667d681d3143a0331b74b6f64416b7e01f44a93a001c38ef74b", "supports": ["summary", "environments", "scope"], "url": "https://arxiv.org/html/2602.01725v1"}], "id": "arxiv:2602.01725", "identifiers": {"arxiv": "2602.01725"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:43Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2602.01725"}], "name": "SafePred: A Predictive Guardrail for Computer-Using Agents via World Models", "notes": "The annotation was reviewed against the complete retained official abstract and the specified sections of the same recorded paper version. Experimental findings and released artifacts have not been independently reproduced.", "published_at": "2026-02-02", "relations": [], "review_status": "curated", "scope": "core", "summary": "SafePred uses world-model risk forecasts to guide action filtering and task replanning. Its desktop and web safety evaluations distinguish policy compliance, task success and success under the specified policy.", "topics": ["safety-security", "world-models", "planning", "evaluation"], "updated_at": "2026-02-02", "url": "https://arxiv.org/abs/2602.01725"}
{"aliases": [], "authors": ["Xilong Wang", "Yinuo Liu", "Zhun Wang", "Dawn Song", "Neil Gong"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "86702018906e64294d7b8daea618c5f152e8b94b604d27d6b164e82aff462564", "submitted_at": "2026-02-03T17:55:04Z", "updated_at": "2026-02-03T17:55:04Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2602.03792v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2602.03792", "identifiers": {"arxiv": "2602.03792"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:43Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2602.03792"}], "name": "WebSentinel: Detecting and Localizing Prompt Injection Attacks for Web Agents", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and released artifacts have not been independently reproduced.", "published_at": "2026-02-03", "relations": [], "review_status": "curated", "scope": "core", "summary": "WebSentinel identifies suspicious webpage segments and checks each against its surrounding page context to detect and locate prompt injections. Its evaluation uses both contaminated and clean webpage datasets.", "topics": ["safety-security", "evaluation", "web-navigation"], "updated_at": "2026-02-03", "url": "https://arxiv.org/abs/2602.03792"}
{"aliases": [], "authors": ["Yuting Ning", "Jaylen Jones", "Zhehao Zhang", "Chentao Ye", "Weitong Ruan", "Junyi Li", "Rahul Gupta", "Huan Sun"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "853f64c020f47129be01778d5212ade8ef78251959ad634ee04ad0fb9fe2989c", "submitted_at": "2026-02-09T18:41:15Z", "updated_at": "2026-06-25T04:36:12Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2602.08995v2"}}, "environments": ["desktop", "web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "kind", "summary", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T11:26:41+00:00", "role": "primary", "sha256": "487a5d21d2385320315eb5a4a2c76e1017973d6027e219be6d31d9328fd5ff9a", "supports": ["summary", "environments", "scope"], "url": "https://arxiv.org/html/2602.08995v2"}], "id": "arxiv:2602.08995", "identifiers": {"arxiv": "2602.08995"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:43Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2602.08995"}], "name": "When Actions Go Off-Task: Detecting and Correcting Misaligned Actions in Computer-Use Agents", "notes": "The annotation was reviewed against the complete retained official abstract and the specified sections of the same recorded paper version. Experimental findings and released artifacts have not been independently reproduced.", "published_at": "2026-02-09", "relations": [], "review_status": "curated", "scope": "core", "summary": "MisActBench provides action-level labels for task alignment, while DeAction checks proposed computer-use actions and requests revisions through structured feedback. The paper separately evaluates offline detection and interactive operation in OSWorld and RedTeamCUA.", "topics": ["safety-security", "evaluation", "robustness"], "updated_at": "2026-06-25", "url": "https://arxiv.org/abs/2602.08995"}
{"aliases": [], "authors": ["Jiacheng Liu", "Yaxin Luo", "Jiacheng Cui", "Xinyi Shang", "Xiaohan Zhao", "Zhiqiang Shen"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "3178684a322fcf07f847473139efdddf0b02213af468f1afbd285549d15bde56", "submitted_at": "2026-02-09T18:55:33Z", "updated_at": "2026-06-21T08:25:39Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2602.09012v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2602.09012", "identifiers": {"arxiv": "2602.09012"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:43Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2602.09012"}], "name": "Next-Gen CAPTCHAs: Leveraging the Cognitive Gap for Scalable and Diverse GUI-Agent Defense", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-02-09", "relations": [], "review_status": "curated", "scope": "core", "summary": "Next-Gen CAPTCHAs introduces generated interactive challenges intended to distinguish human users from GUI agents. The benchmark targets perception, memory, decisions and actions, with backend-supported challenge types designed for scalable instance generation.", "topics": ["evaluation", "safety-security", "data-generation"], "updated_at": "2026-06-21", "url": "https://arxiv.org/abs/2602.09012"}
{"aliases": [], "authors": ["Lepeng Zhao", "Zhenhua Zou", "Shuo Li", "Zhuotao Liu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "727279bf83a67f66fa8f173d028936fe0bdae63e9f23800d98476da10201ad0d", "submitted_at": "2026-02-08T15:50:04Z", "updated_at": "2026-04-26T01:34:23Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2602.10139v3"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2602.10139", "identifiers": {"arxiv": "2602.10139"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:43Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2602.10139"}], "name": "Anonymization-Enhanced Privacy Protection for Mobile GUI Agents: Available but Invisible", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-02-08", "relations": [], "review_status": "curated", "scope": "core", "summary": "An anonymization framework replaces sensitive screen content and instructions with consistent, type-preserving placeholders while mediating actions through a local interaction proxy. Experiments on AndroidLab and PrivScreen examine the privacy–utility trade-off when mobile GUI agents operate over these transformed interfaces.", "topics": ["safety-security", "screen-understanding", "mobile-navigation"], "updated_at": "2026-04-26", "url": "https://arxiv.org/abs/2602.10139"}
{"aliases": [], "authors": ["Renjun Xu", "Yang Yan"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "f0edc70265a8d46a7c22a9c6384de8aa19f79dbb6987f8c23f4a260647a6a3f0", "submitted_at": "2026-02-12T21:33:25Z", "updated_at": "2026-06-02T16:14:54Z", "version": 4, "versioned_url": "https://arxiv.org/abs/2602.12430v4"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-17T12:31:12Z", "role": "primary", "sha256": "c8c95278aa5dcc6a8f2eda3f3b44d61ec1f35adcb6a4c6c9ab328691a1a4f102", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2601.13462%2C2601.14232%2C2601.15331%2C2601.17548%2C2601.18418%2C2601.18491%2C2601.18827%2C2601.20171%2C2601.20613%2C2601.20742%2C2601.21233%2C2601.21253%2C2601.22037%2C2601.22154%2C2601.22511%2C2601.22607%2C2602.01550%2C2602.03117%2C2602.03361%2C2602.05665%2C2602.05746%2C2602.05877%2C2602.06052%2C2602.07398%2C2602.07458%2C2602.07918%2C2602.09007%2C2602.09463%2C2602.09514%2C2602.10122%2C2602.10525%2C2602.11224%2C2602.12268%2C2602.12430%2C2602.13318%2C2602.13543%2C2602.13692%2C2602.14038%2C2602.14337%2C2602.14364%2C2602.16313%2C2602.16317%2C2602.16666%2C2602.16901%2C2602.16930%2C2602.17547%2C2602.17646%2C2602.18846%2C2602.19320%2C2602.19594&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T12:31:12Z", "role": "primary", "sha256": "c8c95278aa5dcc6a8f2eda3f3b44d61ec1f35adcb6a4c6c9ab328691a1a4f102", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?id_list=2601.13462%2C2601.14232%2C2601.15331%2C2601.17548%2C2601.18418%2C2601.18491%2C2601.18827%2C2601.20171%2C2601.20613%2C2601.20742%2C2601.21233%2C2601.21253%2C2601.22037%2C2601.22154%2C2601.22511%2C2601.22607%2C2602.01550%2C2602.03117%2C2602.03361%2C2602.05665%2C2602.05746%2C2602.05877%2C2602.06052%2C2602.07398%2C2602.07458%2C2602.07918%2C2602.09007%2C2602.09463%2C2602.09514%2C2602.10122%2C2602.10525%2C2602.11224%2C2602.12268%2C2602.12430%2C2602.13318%2C2602.13543%2C2602.13692%2C2602.14038%2C2602.14337%2C2602.14364%2C2602.16313%2C2602.16317%2C2602.16666%2C2602.16901%2C2602.16930%2C2602.17547%2C2602.17646%2C2602.18846%2C2602.19320%2C2602.19594&max_results=50"}], "id": "arxiv:2602.12430", "identifiers": {"arxiv": "2602.12430"}, "kind": "survey", "last_verified_at": "2026-10-17T12:31:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2602.12430"}], "name": "Agent Skills for Large Language Models: Architecture, Acquisition, Security, and the Path Forward", "notes": "Reviewed against the complete retained official abstract. Experimental findings and release availability have not been independently reproduced or verified. Adjacent survey of the skill layer: CUA deployment and GUI grounding are explicit parts of the review, but its overall scope extends well beyond interface agents.", "published_at": "2026-02-12", "relations": [], "review_status": "curated", "scope": "adjacent", "summary": "This survey organizes agent skills by architecture, acquisition, deployment and security, including their role in computer-use systems and GUI grounding. It also proposes a permission and lifecycle governance framework for skill deployment.", "topics": ["survey", "tool-integration", "safety-security", "planning"], "updated_at": "2026-06-02", "url": "https://arxiv.org/abs/2602.12430"}
{"aliases": [], "authors": ["Jaechul Roh", "Eugene Bagdasarian", "Hamed Haddadi", "Ali Shahin Shamsabadi"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "fb048b4e5d1d244d689957f30cdcb93cf4dbb750b24ae527b62ecfb04b554eb4", "submitted_at": "2026-02-13T23:02:50Z", "updated_at": "2026-02-13T23:02:50Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2602.13516v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2602.13516", "identifiers": {"arxiv": "2602.13516"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2602.13516"}], "name": "SPILLage: Agentic Oversharing on the Web", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-02-13", "relations": [], "review_status": "curated", "scope": "core", "summary": "Formalizes and measures natural agentic oversharing when browser agents handle private user emails and calendars across live websites.", "topics": ["safety-security", "web-navigation", "evaluation"], "updated_at": "2026-02-13", "url": "https://arxiv.org/abs/2602.13516"}
{"aliases": [], "authors": ["Linxi Jiang", "Zhijie Liu", "Haotian Luo", "Zhiqiang Lin"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "d3b24ac636d86124a70b44e84657b963db4c5af3367ae70e5c35e0bebb3f600f", "submitted_at": "2026-02-28T05:25:03Z", "updated_at": "2026-02-28T05:25:03Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2603.00476v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2603.00476", "identifiers": {"arxiv": "2603.00476"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2603.00476"}], "name": "Atomicity for Agents: Exposing, Exploiting, and Mitigating TOCTOU Vulnerabilities in Browser-Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-02-28", "relations": [], "review_status": "curated", "scope": "core", "summary": "Identifies time-of-check to time-of-use (TOCTOU) race vulnerabilities when web pages mutate between agent planning and action execution.", "topics": ["safety-security", "web-navigation", "state-tracking"], "updated_at": "2026-02-28", "url": "https://arxiv.org/abs/2603.00476"}
{"aliases": [], "authors": ["Haoyu Liu", "Dingcheng Li", "Lukas Rutishauser", "Zeyu Zheng"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "971770e61bf0af91ad1f17b197f09fc166b26bfb4953d3f2c74d457a87e25f3a", "submitted_at": "2026-03-04T18:29:54Z", "updated_at": "2026-10-05T23:01:49Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2603.04364v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2603.04364", "identifiers": {"arxiv": "2603.04364"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:43Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2603.04364"}], "name": "Dual-Modality Multi-Stage Adversarial Safety Training: Robustifying Multimodal Web Agents Against Cross-Modal Attacks", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2026-03-04", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies webpage injections that affect both screenshots and accessibility trees, then trains web agents through teacher imitation, supervised adversarial examples, and adversarial reinforcement learning. The evaluation tests defenses on MiniWoB++ and held-out tasks.", "topics": ["safety-security", "reinforcement-learning", "evaluation"], "updated_at": "2026-10-05", "url": "https://arxiv.org/abs/2603.04364"}
{"aliases": [], "authors": ["Xunzhuo Liu", "Bowei He", "Xue Liu", "Andy Luo", "Haichen Zhang", "Huamin Chen"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "5c7cb0e4784c2a58feeca27a1687d1f73880d9713c16387f0bbff9ed34c427c7", "submitted_at": "2026-03-16T01:31:32Z", "updated_at": "2026-03-16T01:31:32Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2603.14707v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:43Z", "role": "primary", "sha256": "c432f419481a105a8ce71174ec4aea23c4f0d36ddf89cfe1d0974fa349f6cb59", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=400&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2603.14707", "identifiers": {"arxiv": "2603.14707"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:43Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2603.14707"}], "name": "Visual Confused Deputy: Exploiting and Defending Perception Failures in Computer-Using Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2026-03-16", "relations": [], "review_status": "curated", "scope": "core", "summary": "Formalizes a visual confused-deputy failure in which a computer agent acts on a misperceived UI object. A separate guardrail checks both the visual target and the agent's action reasoning against deployment-specific knowledge before permitting execution.", "topics": ["safety-security", "grounding", "robustness"], "updated_at": "2026-03-16", "url": "https://arxiv.org/abs/2603.14707"}
{"aliases": [], "authors": ["Mateusz Dziemian", "Maxwell Lin", "Xiaohan Fu", "Micha Nowak", "Nick Winter", "Eliot Jones", "Andy Zou", "Lama Ahmad", "Kamalika Chaudhuri", "Sahana Chennabasappa", "Xander Davies", "Lauren Deason", "Benjamin L. Edelman", "Tanner Emek", "Ivan Evtimov", "Jim Gust", "Maia Hamin", "Kat He", "Klaudia Krawiecka", "Riccardo Patana", "Neil Perry", "Troy Peterson", "Xiangyu Qi", "Javier Rando", "Zifan Wang", "Zihan Wang", "Spencer Whitman", "Eric Winsor", "Arman Zharmagambetov", "Matt Fredrikson", "Zico Kolter"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e7fe52247e01f331021013dbc3003b3b828e134d7706ec1f587210ad286eee1d", "submitted_at": "2026-03-16T14:49:36Z", "updated_at": "2026-03-16T14:49:36Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2603.15714v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-17T12:31:16Z", "role": "primary", "sha256": "9289bc047b59e93f75703b72f5e92d9277e115238427519c343df550220e351b", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2602.19870%2C2602.22769%2C2602.23166%2C2602.23320%2C2603.00198%2C2603.00842%2C2603.00991%2C2603.01050%2C2603.01145%2C2603.02236%2C2603.02795%2C2603.03278%2C2603.04549%2C2603.06713%2C2603.07106%2C2603.07670%2C2603.07853%2C2603.08706%2C2603.09716%2C2603.10002%2C2603.10664%2C2603.11139%2C2603.11320%2C2603.12483%2C2603.13026%2C2603.13247%2C2603.15714%2C2603.16065%2C2603.16289%2C2603.16839%2C2603.17216%2C2603.17378%2C2603.19313%2C2603.19685%2C2603.21489%2C2603.22527%2C2603.23638%2C2603.24157%2C2603.24639%2C2603.27646%2C2603.28069%2C2603.28407%2C2603.29418%2C2603.29847%2C2604.00528%2C2604.00886%2C2604.01194%2C2604.01438%2C2604.01560%2C2604.02985&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T12:31:16Z", "role": "primary", "sha256": "9289bc047b59e93f75703b72f5e92d9277e115238427519c343df550220e351b", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?id_list=2602.19870%2C2602.22769%2C2602.23166%2C2602.23320%2C2603.00198%2C2603.00842%2C2603.00991%2C2603.01050%2C2603.01145%2C2603.02236%2C2603.02795%2C2603.03278%2C2603.04549%2C2603.06713%2C2603.07106%2C2603.07670%2C2603.07853%2C2603.08706%2C2603.09716%2C2603.10002%2C2603.10664%2C2603.11139%2C2603.11320%2C2603.12483%2C2603.13026%2C2603.13247%2C2603.15714%2C2603.16065%2C2603.16289%2C2603.16839%2C2603.17216%2C2603.17378%2C2603.19313%2C2603.19685%2C2603.21489%2C2603.22527%2C2603.23638%2C2603.24157%2C2603.24639%2C2603.27646%2C2603.28069%2C2603.28407%2C2603.29418%2C2603.29847%2C2604.00528%2C2604.00886%2C2604.01194%2C2604.01438%2C2604.01560%2C2604.02985&max_results=50"}], "id": "arxiv:2603.15714", "identifiers": {"arxiv": "2603.15714"}, "kind": "paper", "last_verified_at": "2026-10-17T12:31:16Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2603.15714"}], "name": "How Vulnerable Are AI Agents to Indirect Prompt Injections? Insights from a Large-Scale Public Competition", "notes": "Reviewed against the complete retained official abstract. Experimental findings and release availability have not been independently reproduced or verified.", "published_at": "2026-03-16", "relations": [], "review_status": "curated", "scope": "core", "summary": "This public red-teaming study evaluates indirect prompt injection that both changes agent behavior and hides the compromise from the final response. Its scenarios include computer use alongside tool calling and coding, with analyses of transfer and model-dependent vulnerability.", "topics": ["safety-security", "evaluation", "robustness"], "updated_at": "2026-03-16", "url": "https://arxiv.org/abs/2603.15714"}
{"aliases": [], "authors": ["Nathan Zhao"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e3b429d277ec776af18977a72aaeb95000a4fedf66aa89c0366ab02120d4cdbe", "submitted_at": "2026-03-18T04:41:16Z", "updated_at": "2026-03-18T04:41:16Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2603.17357v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2603.17357", "identifiers": {"arxiv": "2603.17357"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:39Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2603.17357"}], "name": "WebPII: Benchmarking Visual PII Detection for Computer-Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-03-18", "relations": [], "review_status": "curated", "scope": "core", "summary": "WebPII provides synthetic e-commerce screenshots with fine-grained annotations for sensitive information, including partially filled forms and transaction identifiers. It also trains WebRedact to detect visual PII as a privacy component for computer-use agents.", "topics": ["evaluation", "data-generation", "safety-security", "screen-understanding"], "updated_at": "2026-03-18", "url": "https://arxiv.org/abs/2603.17357"}
{"aliases": [], "authors": ["Haochen Zhao", "Shaoyang Cui"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "0beb3fe342f080f0727181aba55fc1167d2512dd04542986448c5006b6b8213b", "submitted_at": "2026-03-19T11:14:45Z", "updated_at": "2026-03-19T11:14:45Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2603.18762v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2603.18762", "identifiers": {"arxiv": "2603.18762"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2603.18762"}], "name": "ClawTrap: A MITM-Based Red-Teaming Framework for Real-World OpenClaw Security Evaluation", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-03-19", "relations": [], "review_status": "curated", "scope": "core", "summary": "Presents a man-in-the-middle network red-teaming framework for evaluating autonomous OpenClaw agents under live traffic manipulation.", "topics": ["safety-security", "evaluation", "web-navigation"], "updated_at": "2026-03-19", "url": "https://arxiv.org/abs/2603.18762"}
{"aliases": [], "authors": ["Yizhe Zhao", "Yongjian Fu", "Zihao Feng", "Hao Pan", "Yongheng Deng", "Yaoxue Zhang", "Ju Ren"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "f25f4999fc9d20433f96e22651311cccf78b5e2ad7cbc5578fe1765ac8be6a00", "submitted_at": "2026-03-20T12:19:41Z", "updated_at": "2026-03-20T12:19:41Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2603.20351v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T04:21:52Z", "role": "primary", "sha256": "c194d15b7e422aae48b4e05216a1f1d84471b6666718b9d3dd4c9d250c356e77", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2210.03347%2C2403.04984%2C2405.07090%2C2406.06282%2C2409.00088%2C2409.10811%2C2411.10943%2C2503.09241%2C2503.20279%2C2503.23350%2C2505.05440%2C2505.10593%2C2505.16422%2C2505.21279%2C2505.24878%2C2506.02153%2C2506.02839%2C2506.05079%2C2506.07976%2C2508.02046%2C2511.07885%2C2511.11248%2C2512.21720%2C2603.20351%2C2604.19299&max_results=25"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T04:21:52Z", "role": "primary", "sha256": "c194d15b7e422aae48b4e05216a1f1d84471b6666718b9d3dd4c9d250c356e77", "supports": ["name", "kind", "scope", "environments", "topics", "summary", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2210.03347%2C2403.04984%2C2405.07090%2C2406.06282%2C2409.00088%2C2409.10811%2C2411.10943%2C2503.09241%2C2503.20279%2C2503.23350%2C2505.05440%2C2505.10593%2C2505.16422%2C2505.21279%2C2505.24878%2C2506.02153%2C2506.02839%2C2506.05079%2C2506.07976%2C2508.02046%2C2511.07885%2C2511.11248%2C2512.21720%2C2603.20351%2C2604.19299&max_results=25"}], "id": "arxiv:2603.20351", "identifiers": {"arxiv": "2603.20351"}, "kind": "paper", "last_verified_at": "2026-10-18T04:21:52Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2603.20351"}], "name": "MANA: Towards Efficient Mobile Ad Detection via Multimodal Agentic UI Navigation", "notes": "Reviewed against the complete official arXiv abstract. Experimental results and linked artifact availability have not been independently reproduced.", "published_at": "2026-03-20", "relations": [], "review_status": "curated", "scope": "core", "summary": "MANA combines visual, temporal and exploration signals to guide navigation toward mobile advertisements. It evaluates ad detection and interaction cost on commercial smartphones, including obfuscated-ad cases.", "topics": ["mobile-navigation", "screen-understanding", "efficiency", "safety-security"], "updated_at": "2026-03-20", "url": "https://arxiv.org/abs/2603.20351"}
{"aliases": [], "authors": ["Yutao Luo", "Haotian Zhu", "Shuchao Pang", "Zhigang Lu", "Tian Dong", "Yongbin Zhou", "Minhui Xue"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "37960f42162d2bff2fca5e71e350f1ef7c4d86e3f438fd777a88dc2280d3aabe", "submitted_at": "2026-03-24T09:51:43Z", "updated_at": "2026-03-24T09:51:43Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2603.23007v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2603.23007", "identifiers": {"arxiv": "2603.23007"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:39Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2603.23007"}], "name": "AgentRAE: Remote Action Execution through Notification-based Visual Backdoors against Screenshots-based Mobile GUI Agents", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2026-03-24", "relations": [], "review_status": "curated", "scope": "core", "summary": "AgentRAE studies notification-based visual backdoors that redirect mobile GUI-agent actions. Its training separates sensitivity to icon differences from trigger/action association, with evaluations of clean behavior and targeted redirection.", "topics": ["safety-security", "robustness", "evaluation"], "updated_at": "2026-03-24", "url": "https://arxiv.org/abs/2603.23007"}
{"aliases": [], "authors": ["Qianlong Lan", "Anuj Kaul"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "bb8164fa076c55cddf73508e6b68ce24a6594917e65862504ba9f08b4e97ee48", "submitted_at": "2026-03-24T23:49:15Z", "updated_at": "2026-03-24T23:49:15Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2603.23791v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2603.23791", "identifiers": {"arxiv": "2603.23791"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2603.23791"}], "name": "The Cognitive Firewall:Securing Browser Based AI Agents Against Indirect Prompt Injection Via Hybrid Edge Cloud Defense", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-03-24", "relations": [], "review_status": "curated", "scope": "core", "summary": "Proposes a split-compute browser defense pairing a local visual sentinel, a cloud planner, and deterministic policy enforcement against indirect prompt injection.", "topics": ["safety-security", "web-navigation"], "updated_at": "2026-03-24", "url": "https://arxiv.org/abs/2603.23791"}
{"aliases": [], "authors": ["Zhengyang Tang", "Ke Ji", "Xidong Wang", "Zihan Ye", "Xinyuan Wang", "Yiduo Guo", "Ziniu Li", "Chenxin Li", "Jingyuan Hu", "Shunian Chen", "Tongxu Luo", "Jiaxi Bi", "Zeyu Qin", "Shaobo Wang", "Xin Lai", "Pengyuan Lyu", "Junyi Li", "Can Xu", "Chengquan Zhang", "Han Hu", "Ming Yan", "Benyou Wang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "3f137028b7608598fb1885a0bf1330e00c9404851571f856665aae1af1d9d63c", "submitted_at": "2026-04-01T14:50:50Z", "updated_at": "2026-04-02T13:51:55Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2604.00986v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2604.00986", "identifiers": {"arxiv": "2604.00986"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2604.00986"}], "name": "Do Phone-Use Agents Respect Your Privacy?", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-04-01", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces MyPhoneBench to verify whether phone-use agents minimize personal data disclosure when filling forms during benign mobile tasks.", "topics": ["safety-security", "mobile-navigation", "evaluation"], "updated_at": "2026-04-02", "url": "https://arxiv.org/abs/2604.00986"}
{"aliases": [], "authors": ["Wei Zou", "Mingwen Dong", "Miguel Romero Calvo", "Shuaichen Chang", "Jiang Guo", "Dongkyu Lee", "Xing Niu", "Xiaofei Ma", "Yanjun Qi", "Jiarong Jiang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "a37f4bdc495e10e4ae301ebff6faa28cdfa6de152d4619c051083b5dce41e6ba", "submitted_at": "2026-04-03T01:25:12Z", "updated_at": "2026-04-07T14:45:15Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2604.02623v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2604.02623", "identifiers": {"arxiv": "2604.02623"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:39Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2604.02623"}], "name": "Poison Once, Exploit Forever: Environment-Injected Memory Poisoning Attacks on Web Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2026-04-03", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies persistent web-agent memory contamination introduced through environmental observations instead of direct memory access. Experiments on WebArena and VisualWebArena examine cross-session effects and how interaction failures alter susceptibility.", "topics": ["safety-security", "memory", "web-navigation"], "updated_at": "2026-04-07", "url": "https://arxiv.org/abs/2604.02623"}
{"aliases": [], "authors": ["Chaoran Chen", "Zhiping Zhang", "Zeya Chen", "Eryue Xu", "Yinuo Yang", "Ibrahim Khalilov", "Simret A Gebreegziabher", "Yanfang Ye", "Ziang Xiao", "Yaxing Yao", "Tianshi Li", "Toby Jia-Jun Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "1411ee71ad8f65e4941af4ef538f52da6e9bf9c57fb537b6ecf37c8478609693", "submitted_at": "2026-04-06T17:57:16Z", "updated_at": "2026-09-12T06:05:52Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2604.04918v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2604.04918", "identifiers": {"arxiv": "2604.04918"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:39Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2604.04918"}], "name": "Comparing Human Oversight Strategies for Computer-Use Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced.", "published_at": "2026-04-06", "relations": [], "review_status": "curated", "scope": "core", "summary": "Compares four human oversight strategies during live web-agent sessions, examining exposure to problematic actions and opportunities to intervene. The analysis identifies timing and attention problems in control handoffs and upfront plan approval.", "topics": ["human-in-the-loop", "safety-security", "evaluation"], "updated_at": "2026-09-12", "url": "https://arxiv.org/abs/2604.04918"}
{"aliases": [], "authors": ["Guruprasad Viswanathan Ramesh", "Asmit Nayak", "Basieem Siddique", "Kassem Fawaz"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "400d8a9f7aed4f150dd5198d9a1561d2df0454069495594ca165ad4254a1be59", "submitted_at": "2026-04-07T18:43:21Z", "updated_at": "2026-06-16T22:37:53Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2604.06367v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2604.06367", "identifiers": {"arxiv": "2604.06367"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:39Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2604.06367"}], "name": "WebSP-Eval: Evaluating Web Agents on Website Security and Privacy Tasks", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2026-04-07", "relations": [], "review_status": "curated", "scope": "core", "summary": "WebSP-Eval evaluates agents carrying out website security and privacy tasks such as preference changes and session management. A browser extension controls account and initial state across runs, enabling analysis of failures on stateful UI controls.", "topics": ["safety-security", "evaluation", "web-navigation", "state-tracking"], "updated_at": "2026-06-16", "url": "https://arxiv.org/abs/2604.06367"}
{"aliases": [], "authors": ["Wenkui Yang", "Chao Jin", "Haisu Zhu", "Weilin Luo", "Derek Yuen", "Kun Shao", "Junxian Duan", "Huaibo Huang", "Jie Cao", "Ran He"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "a5427eedff0e97f9c73101403754c38e8928ad71abab7b7eff5f136a30ee2205", "submitted_at": "2026-04-09T05:32:34Z", "updated_at": "2026-07-08T02:39:35Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2604.07831v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2604.07831", "identifiers": {"arxiv": "2604.07831"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:39Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2604.07831"}], "name": "Are GUI Agents Focused Enough? Automated Distraction via Semantic-level UI Element Injection", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2026-04-09", "relations": [], "review_status": "curated", "scope": "core", "summary": "This study tests black-box GUI-grounding failures caused by strategically added, innocuous-looking UI elements. Its iterative overlay search is compared with random injection and repeated independent grounding trials to distinguish persistent misdirection from incidental clutter.", "topics": ["safety-security", "grounding", "robustness", "evaluation"], "updated_at": "2026-07-08", "url": "https://arxiv.org/abs/2604.07831"}
{"aliases": [], "authors": ["Yushi Feng", "Junye Du", "Qifan Wang", "Zizhan Ma", "Qian Niu", "Yutaka Matsuo", "Long Feng", "Lequan Yu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e53647528b9ee9d98f93ee665d0d886f9b400a129873e240704041df291f5749", "submitted_at": "2026-04-10T09:41:21Z", "updated_at": "2026-04-10T09:41:21Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2604.09155v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:32:00Z", "role": "primary", "sha256": "4d09ebfb093e43a4c44adef86c8fa8374f90d907fc33bd8e12a1315e62d5a583", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2603.12823%2C2603.14707%2C2603.15039%2C2603.17357%2C2603.18429%2C2603.19191%2C2603.24440%2C2603.25226%2C2603.25864%2C2603.26211%2C2603.26648%2C2603.29318%2C2604.01535%2C2604.02947%2C2604.04399%2C2604.05477%2C2604.06126%2C2604.06182%2C2604.06367%2C2604.07429%2C2604.08455%2C2604.08516%2C2604.08523%2C2604.09155%2C2604.09574%2C2604.09587%2C2604.09815%2C2604.09937%2C2604.11259%2C2604.11784%2C2604.12284%2C2604.13019%2C2604.13822%2C2604.14113%2C2604.14262%2C2604.14956%2C2604.15093%2C2604.15309%2C2604.17817%2C2604.17821%2C2604.18860%2C2604.24964%2C2604.25380%2C2604.25562%2C2604.27151%2C2604.27253%2C2604.27776%2C2605.00642%2C2605.06393%2C2605.06664&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:32:00Z", "role": "primary", "sha256": "4d09ebfb093e43a4c44adef86c8fa8374f90d907fc33bd8e12a1315e62d5a583", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2603.12823%2C2603.14707%2C2603.15039%2C2603.17357%2C2603.18429%2C2603.19191%2C2603.24440%2C2603.25226%2C2603.25864%2C2603.26211%2C2603.26648%2C2603.29318%2C2604.01535%2C2604.02947%2C2604.04399%2C2604.05477%2C2604.06126%2C2604.06182%2C2604.06367%2C2604.07429%2C2604.08455%2C2604.08516%2C2604.08523%2C2604.09155%2C2604.09574%2C2604.09587%2C2604.09815%2C2604.09937%2C2604.11259%2C2604.11784%2C2604.12284%2C2604.13019%2C2604.13822%2C2604.14113%2C2604.14262%2C2604.14956%2C2604.15093%2C2604.15309%2C2604.17817%2C2604.17821%2C2604.18860%2C2604.24964%2C2604.25380%2C2604.25562%2C2604.27151%2C2604.27253%2C2604.27776%2C2605.00642%2C2605.06393%2C2605.06664&max_results=50"}], "id": "arxiv:2604.09155", "identifiers": {"arxiv": "2604.09155"}, "kind": "paper", "last_verified_at": "2026-10-16T06:32:00Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2604.09155"}], "name": "CORA: Conformal Risk-Controlled Agents for Safeguarded Mobile GUI Automation", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experimental results and artifact availability have not been independently reproduced. The annotation describes risk calibration without asserting an unconditional safety guarantee.", "published_at": "2026-04-10", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces CORA, which calibrates a risk model to choose between executing and abstaining from mobile GUI actions, then diagnoses rejected actions to suggest interventions. Phone-Harm provides step-level safety labels for evaluating this safeguarding process.", "topics": ["safety-security", "mobile-navigation", "evaluation"], "updated_at": "2026-04-10", "url": "https://arxiv.org/abs/2604.09155"}
{"aliases": [], "authors": ["Yulin Chen", "Tri Cao", "Haoran Li", "Yue Liu", "Yibo Li", "Yufei He", "Le Minh Khoi", "Yangqiu Song", "Shuicheng Yan", "Bryan Hooi"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "64d4323050a5ad0f7265231337eb22a4c0823cee41587099c8bdc5db6b320dea", "submitted_at": "2026-04-14T04:50:35Z", "updated_at": "2026-04-14T04:50:35Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2604.12284v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2604.12284", "identifiers": {"arxiv": "2604.12284"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:39Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2604.12284"}], "name": "WebAgentGuard: A Reasoning-Driven Guard Model for Detecting Prompt Injection Attacks in Web Agents", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2026-04-14", "relations": [], "review_status": "curated", "scope": "core", "summary": "WebAgentGuard separates prompt-injection detection from the operating web agent through a parallel multimodal guard. The guard is trained on synthetic visual/textual examples using reasoning-focused supervised fine-tuning followed by reinforcement learning.", "topics": ["safety-security", "data-generation", "imitation-learning", "reinforcement-learning", "evaluation"], "updated_at": "2026-04-14", "url": "https://arxiv.org/abs/2604.12284"}
{"aliases": [], "authors": ["Jiachen Qian", "Zhaolu Kang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "64e47028e752dd023e71414664a193beba74e1c4f1afe765fdf992d560dbea27", "submitted_at": "2026-04-15T10:46:38Z", "updated_at": "2026-04-15T10:46:38Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2604.16515v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2604.16515", "identifiers": {"arxiv": "2604.16515"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2604.16515"}], "name": "Penny Wise, Pixel Foolish: Bypassing Price Constraints in Multimodal Agents via Visual Adversarial Perturbations", "notes": "Scope is screenshot-based interface understanding and safety. The retained abstract does not establish a specific operating system or live transaction execution; no attack results were independently reproduced.", "published_at": "2026-04-15", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces PriceBlind, a visual perturbation method for controlled screenshot-based evaluation of price-constrained multimodal agents. The reported transfer experiments use a simplified single-turn coordinate-selection protocol.", "topics": ["safety-security", "robustness"], "updated_at": "2026-04-15", "url": "https://arxiv.org/abs/2604.16515"}
{"aliases": [], "authors": ["Zehao Lin", "Xixuan Hao", "Renyu Fu", "Shaobo Cui", "Kai Chen", "Chunyu Li", "Zhiyu Li", "Feiyu Xiong"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "fb75ea30086ecf069d8adda1513a6e906cfcfcedcd4cab01d0986a850aa8009a", "submitted_at": "2026-04-17T06:28:22Z", "updated_at": "2026-09-22T07:33:24Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2604.16548v3"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T13:00:08.054642+00:00", "role": "primary", "sha256": "f795535d55a4def0f2ce7dfc41a9d1d22ad38eddf48fb4f1feed98b6b5b66721", "supports": ["summary", "scope", "notes"], "url": "https://arxiv.org/html/2604.16548v3"}], "id": "arxiv:2604.16548", "identifiers": {"arxiv": "2604.16548"}, "kind": "survey", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2604.16548"}], "name": "A Survey on Long-Term Memory Security in LLM Agents: Attacks, Defenses, and Governance Across the Memory Lifecycle", "notes": "Adjacent memory-security survey. Full-text Sections 3.1 and 3.3 discuss poisoned web observations persisted into later agent actions; the work is broader than GUI agents and does not establish a platform implementation.", "published_at": "2026-04-17", "relations": [], "review_status": "curated", "scope": "adjacent", "summary": "Organizes security risks and defenses for persistent agent memory into six lifecycle phases and proposes five verifiable governance primitives.", "topics": ["memory", "safety-security", "survey"], "updated_at": "2026-09-22", "url": "https://arxiv.org/abs/2604.16548"}
{"aliases": [], "authors": ["Wenpeng Xu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "8eb499729268ef287a9b9bf73275f526d47dabf2d7de399beaead2fe54cfd74e", "submitted_at": "2026-04-20T21:36:16Z", "updated_at": "2026-04-20T21:36:16Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2604.18860v1"}}, "environments": ["desktop", "web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2604.18860", "identifiers": {"arxiv": "2604.18860"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:39Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2604.18860"}], "name": "Temporal UI State Inconsistency in Desktop GUI Agents: Formalizing and Defending Against TOCTOU Attacks on Computer-Use Agents", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2026-04-20", "relations": [], "review_status": "curated", "scope": "core", "summary": "This study examines UI changes between a desktop agent’s screenshot observation and action dispatch. Its pre-execution state checks address overlay and focus manipulation, while the reported DOM-injection blind spot limits the defense’s coverage.", "topics": ["safety-security", "robustness", "evaluation", "state-tracking"], "updated_at": "2026-04-20", "url": "https://arxiv.org/abs/2604.18860"}
{"aliases": [], "authors": ["Zheng Wu", "Yi Hua", "Zhaoyuan Huang", "Chenhao Xue", "Yijie Lu", "Pengzhou Cheng", "Zongru Wu", "Lingzhong Dong", "Gongshen Liu", "Xinghao Jiang", "Zhuosheng Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "772420683208561df8493c9ed5d68b341e9e098f5729945b972f3555e976d799", "submitted_at": "2026-04-27T11:44:26Z", "updated_at": "2026-04-27T11:44:26Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2604.24348v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2604.24348", "identifiers": {"arxiv": "2604.24348"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2604.24348"}], "name": "OS-SPEAR: A Toolkit for the Safety, Performance,Efficiency, and Robustness Analysis of OS Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-04-27", "relations": [], "review_status": "curated", "scope": "core", "summary": "Provides an evaluation toolkit for joint safety, efficiency, and multimodal robustness analysis of operating-system GUI agents.", "topics": ["evaluation", "safety-security", "efficiency", "robustness"], "updated_at": "2026-04-27", "url": "https://arxiv.org/abs/2604.24348"}
{"aliases": [], "authors": ["Mengyao Du", "Han Fang", "Haokai Ma", "Jiahao Chen", "Kai Xu", "Quanjun Yin", "Ee-Chien Chang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "d2610c2658221d7da0bd888bf7b143ca6c4175791058570f6562cfa0c8c3a946", "submitted_at": "2026-04-28T12:32:21Z", "updated_at": "2026-04-28T12:32:21Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2604.25562v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2604.25562", "identifiers": {"arxiv": "2604.25562"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:39Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2604.25562"}], "name": "SnapGuard: Lightweight Prompt Injection Detection for Screenshot-Based Web Agents", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2026-04-28", "relations": [], "review_status": "curated", "scope": "core", "summary": "SnapGuard detects screenshot-based prompt injection through complementary visual-stability and recovered textual cues. Its evaluation compares detection quality and inference cost across attack and benign-page settings, rather than establishing universal runtime prevention.", "topics": ["safety-security", "screen-understanding", "evaluation", "efficiency"], "updated_at": "2026-04-28", "url": "https://arxiv.org/abs/2604.25562"}
{"aliases": [], "authors": ["Sohom Datta", "Alex Nahapetyan", "William Enck", "Alexandros Kapravelos"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "0ff44befcfdb97b67d1680c97784f9e9665d8d9879c093984f42d06c9516a342", "submitted_at": "2026-05-06T23:19:38Z", "updated_at": "2026-05-06T23:19:38Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2605.05509v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2605.05509", "identifiers": {"arxiv": "2605.05509"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2605.05509"}], "name": "WAAA! Web Adversaries Against Agentic Browsers", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-05-06", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates agentic browsers against classical web attacks and visual social-engineering lures beyond indirect prompt injection.", "topics": ["safety-security", "web-navigation", "evaluation"], "updated_at": "2026-05-06", "url": "https://arxiv.org/abs/2605.05509"}
{"aliases": [], "authors": ["Yilin Zhang", "Yingkai Hua", "Chunyu Wei", "Xin Wang", "Yueguo Chen"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "b57d784c70b50d2c25055c6d139c0a5acfc97da3a4effbd63536b72d6df38a69", "submitted_at": "2026-05-10T12:11:52Z", "updated_at": "2026-05-10T12:11:52Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2605.09497v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2605.09497", "identifiers": {"arxiv": "2605.09497"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:39Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2605.09497"}], "name": "Don't Click That: Teaching Web Agents to Resist Deceptive Interfaces", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-05-10", "relations": [], "review_status": "curated", "scope": "core", "summary": "DUDE combines asymmetric hybrid-reward learning with summaries of deceptive-interface failures to guide safer web actions. Its RUC benchmark evaluates deception susceptibility alongside ordinary task performance across interface scenarios.", "topics": ["safety-security", "reinforcement-learning", "evaluation", "robustness"], "updated_at": "2026-05-10", "url": "https://arxiv.org/abs/2605.09497"}
{"aliases": [], "authors": ["Julien Piet", "Annabella Chow", "Yiwei Hou", "Muxi Lyu", "Sylvie Venuto", "Jinhao Zhu", "Raluca Ada Popa", "David Wagner"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "572be49190554bbfb6accac4c87ef91608f88e90d06250f229db2922e19fcd56", "submitted_at": "2026-05-14T02:48:57Z", "updated_at": "2026-05-14T02:48:57Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2605.14290v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:39Z", "role": "primary", "sha256": "27c33576ed43c7bc63d4b290f6558553bd1d3aa112a7358138f4045e1bc14918", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=300&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2605.14290", "identifiers": {"arxiv": "2605.14290"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:39Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2605.14290"}], "name": "Web Agents Should Adopt the Plan-Then-Execute Paradigm", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced. Directly analyzes browser-agent prompt-injection boundaries and WebArena task structure, but advocates typed website operations rather than demonstrating a new GUI-control policy.", "published_at": "2026-05-14", "relations": [], "review_status": "curated", "scope": "adjacent", "summary": "Argues for compiling web tasks into predefined execution graphs before exposing an agent to untrusted runtime content. Its WebArena task analysis identifies typed, semantically complete website tools as a prerequisite and distinguishes this interface proposal from reactive click-level control.", "topics": ["planning", "safety-security", "tool-integration"], "updated_at": "2026-05-14", "url": "https://arxiv.org/abs/2605.14290"}
{"aliases": [], "authors": ["Yutong Huang", "Vikranth Srivatsa", "Alex Asch", "Hansin Tushar Patwa", "Yiying Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "51576fb9bde23f769d3200a0d593ad610eb27248917132aff1bcf6eb7ff03cea", "submitted_at": "2026-05-17T08:24:01Z", "updated_at": "2026-05-17T08:24:01Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2605.17320v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2605.17320", "identifiers": {"arxiv": "2605.17320"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2605.17320"}], "name": "TClone: Low-Latency Forking of Live GUI Environments for Computer-Use Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-05-17", "relations": [], "review_status": "curated", "scope": "core", "summary": "TClone versions live GUI workspaces through snapshots, isolated forks, rollback and selective commit or merge. It separates fast branch creation from durable checkpoints using copy-on-write sharing, filesystem versioning and asynchronous checkpointing, with end-to-end agent-loop latency measurements.", "topics": ["sandbox", "efficiency", "safety-security", "automation"], "updated_at": "2026-05-17", "url": "https://arxiv.org/abs/2605.17320"}
{"aliases": [], "authors": ["Ruoqi Guo", "Yi Liu", "Gelei Deng", "Yiheng Xiong", "Yuekang Li", "Ying Zhang", "Leo Yu Zhang", "Lida Zhao", "Ji Jie", "Yuxiao Lu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "4f39433695b9574b275fc844ca6b10a43baf66aed57048bf841f41bf08d4ac9a", "submitted_at": "2026-05-27T08:09:53Z", "updated_at": "2026-05-27T08:09:53Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2605.28116v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2605.28116", "identifiers": {"arxiv": "2605.28116"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2605.28116"}], "name": "MIRAGE: Context-Aware Prompt Injection against Mobile GUI Agents via User-Generated Content", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-05-27", "relations": [], "review_status": "curated", "scope": "core", "summary": "MIRAGE constructs mobile screenshot injections in ordinary user-content regions through localization, contextual payload generation and realism curation. Its evaluation separates visual realism from attack effectiveness against pixel-observing GUI agents.", "topics": ["safety-security", "data-generation", "evaluation", "robustness"], "updated_at": "2026-05-27", "url": "https://arxiv.org/abs/2605.28116"}
{"aliases": [], "authors": ["Yanqiu Zhao", "Dongying Zheng", "Kaibo Huang", "Yukun Wei", "Zhongliang Yang", "Linna Zhou"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "6eb9b67f7e51edecfbe57af47b6bad445c05129c11c377d32ea84369b430f895", "submitted_at": "2026-05-27T15:51:22Z", "updated_at": "2026-05-29T04:03:46Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2605.28646v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2605.28646", "identifiers": {"arxiv": "2605.28646"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2605.28646"}], "name": "MaskClaw: Edge-Side Personalized Privacy Arbitration for GUI Agents with Behavior-Driven Skill Evolution", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-05-27", "relations": [], "review_status": "curated", "scope": "core", "summary": "MaskClaw makes local Allow, Mask or Ask decisions before GUI screenshots leave a trusted environment, using visual evidence and personalized policy memory. Its benchmark and designed evolution scenarios test how corrections and edits can become privacy skills checked by a sandbox gate.", "topics": ["safety-security", "memory", "human-in-the-loop", "evaluation"], "updated_at": "2026-05-29", "url": "https://arxiv.org/abs/2605.28646"}
{"aliases": [], "authors": ["Jia Yu", "Zilong Wang", "Xinyang Jiang", "Dongsheng Li", "Shuo Wang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "b23c12cafdd2d4a3ef63027c914ed3894c61c555edd89468c56ae40c3ec6cea9", "submitted_at": "2026-06-02T06:02:35Z", "updated_at": "2026-06-02T06:02:35Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2606.03203v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.03203", "identifiers": {"arxiv": "2606.03203"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.03203"}], "name": "MedCUA-Bench: A Screenshot-Only Benchmark for Clinical Computer-Use Agents", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2026-06-02", "relations": [], "review_status": "curated", "scope": "core", "summary": "MedCUA-Bench evaluates screenshot-only clinical interface workflows using paired intent- and step-level goals. It combines reconstructed medical interfaces with completion and clinical-safety checks, and separately reports results on a real OpenEMR system.", "topics": ["evaluation", "safety-security", "science", "workflow"], "updated_at": "2026-06-02", "url": "https://arxiv.org/abs/2606.03203"}
{"aliases": [], "authors": ["Nicholas Saban"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "7b649c0cef63bc15558aa4498a65043644c109faaf6403b610c2132fd3f5706d", "submitted_at": "2026-06-03T01:21:59Z", "updated_at": "2026-06-03T01:21:59Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2606.05233v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.05233", "identifiers": {"arxiv": "2606.05233"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.05233"}], "name": "Domain-Conditioned Safety in Frontier Computer-Using Agents: A 793-Episode Browser Benchmark, a Coding-Domain Cross-Reference, and a Reproducibility Audit of Recent Red-Teaming", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-06-03", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces CUA-HandCrafted, a browser-agent red-teaming benchmark that varies attack templates, tasks and system prompts. Its domain comparison and reproducibility audit caution against transferring browser-surface safety conclusions to coding agents or unreleased optimized injections.", "topics": ["safety-security", "evaluation", "robustness"], "updated_at": "2026-06-03", "url": "https://arxiv.org/abs/2606.05233"}
{"aliases": [], "authors": ["Siyu Shen", "Fenghao Xu", "Wenrui Diao", "Kehuan Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "fed1c6e109cc5d34aebe3bb333955e8bf792264d4c1ed0893d76534a77649afd", "submitted_at": "2026-06-10T20:48:39Z", "updated_at": "2026-06-16T03:15:00Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2606.12666v2"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.12666", "identifiers": {"arxiv": "2606.12666"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.12666"}], "name": "CAPED: Context-Aware Privacy Exposure Defense for Mobile GUI Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-06-10", "relations": [], "review_status": "curated", "scope": "core", "summary": "CAPED filters mobile screenshots before upload by combining task requirements, screen context and parsed UI elements to mask incidental private content. AndroidWorld and a controlled seeded privacy evaluation examine leakage reduction alongside task utility and prototype costs.", "topics": ["safety-security", "screen-understanding", "mobile-navigation"], "updated_at": "2026-06-16", "url": "https://arxiv.org/abs/2606.12666"}
{"aliases": [], "authors": ["Zihao Wang", "Yiming Li", "Yutong Wu", "Kangjie Chen", "Zheyu Liu", "Fok Kar Wai", "Pin-Yu Chen", "Vrizlynn L. L. Thing", "Bo Li", "Dacheng Tao", "Tianwei Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "88b7e2ee8717661e4f8b417916e35b844171ecd8a8a2add657cf341c8f759536", "submitted_at": "2026-06-11T14:12:43Z", "updated_at": "2026-07-27T14:19:24Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2606.13385v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.13385", "identifiers": {"arxiv": "2606.13385"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.13385"}], "name": "Who Pays the Price? Stakeholder-Centric Prompt Injection Benchmarking for Real-world Web Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-06-11", "relations": [], "review_status": "curated", "scope": "core", "summary": "StakeBench evaluates prompt-injection harm to different stakeholders in shopping web-agent workflows. Its executable cases separate attack objectives and assess both outcomes and execution processes to reveal how the same attack pattern can affect different parties.", "topics": ["evaluation", "safety-security", "web-navigation"], "updated_at": "2026-07-27", "url": "https://arxiv.org/abs/2606.13385"}
{"aliases": [], "authors": ["Mina Mohammadmirzaei", "Jeffrey Flanigan"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "7bf0297fba942441b19bcfeba147838bdef888b428f1150c2462d8bf3d986882", "submitted_at": "2026-06-13T00:32:24Z", "updated_at": "2026-10-09T19:45:15Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2606.15034v3"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.15034", "identifiers": {"arxiv": "2606.15034"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.15034"}], "name": "OSGuard: A Benchmark for Safety in Computer-Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-06-13", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates action-level guardrails and full computer tasks with environment-dependent safety constraints, separating safe completion from nominal success.", "topics": ["safety-security", "evaluation"], "updated_at": "2026-10-09", "url": "https://arxiv.org/abs/2606.15034"}
{"aliases": [], "authors": ["Y. H. Zhou", "Z. M. Ma", "Y. J. Zhou", "Y. T. Li", "H. X. Xiang", "Y. M. Cheng", "T. L. Chen", "K. J. Zhang", "Z. H. Nan", "J. H. Ni", "Z. Wu", "Q. Y. Pan", "S. Zhang", "S. Cheng", "M. Y. Luo"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "6311b22ddb2262c74efbe79acc8856335f6870aa24197c507f3f7b048d6f7901", "submitted_at": "2026-06-15T12:53:55Z", "updated_at": "2026-06-15T12:53:55Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2606.16659v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:33:05Z", "role": "primary", "sha256": "2e5722527f8da4b2cf06372d1e2daccd33ef7fdc7151704b411a0bddd734beae", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2606.10423%2C2606.10522%2C2606.10742%2C2606.11078%2C2606.11520%2C2606.12485%2C2606.12666%2C2606.13239%2C2606.13385%2C2606.14106%2C2606.14832%2C2606.15017%2C2606.15034%2C2606.16659%2C2606.16748%2C2606.16802%2C2606.17321%2C2606.17645%2C2606.17929%2C2606.18448%2C2606.18671%2C2606.18890%2C2606.19388%2C2606.19926%2C2606.19930%2C2606.20363%2C2606.20636%2C2606.20717%2C2606.20724%2C2606.20785%2C2606.21654%2C2606.22864%2C2606.22948%2C2606.23449%2C2606.24515%2C2606.24525%2C2606.25705%2C2606.25760%2C2606.27330%2C2606.27595%2C2606.29445%2C2606.29537%2C2606.29705%2C2606.30697%2C2606.31154%2C2606.31270%2C2606.31410%2C2606.31612%2C2607.00502%2C2607.04425&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:33:05Z", "role": "primary", "sha256": "2e5722527f8da4b2cf06372d1e2daccd33ef7fdc7151704b411a0bddd734beae", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2606.10423%2C2606.10522%2C2606.10742%2C2606.11078%2C2606.11520%2C2606.12485%2C2606.12666%2C2606.13239%2C2606.13385%2C2606.14106%2C2606.14832%2C2606.15017%2C2606.15034%2C2606.16659%2C2606.16748%2C2606.16802%2C2606.17321%2C2606.17645%2C2606.17929%2C2606.18448%2C2606.18671%2C2606.18890%2C2606.19388%2C2606.19926%2C2606.19930%2C2606.20363%2C2606.20636%2C2606.20717%2C2606.20724%2C2606.20785%2C2606.21654%2C2606.22864%2C2606.22948%2C2606.23449%2C2606.24515%2C2606.24525%2C2606.25705%2C2606.25760%2C2606.27330%2C2606.27595%2C2606.29445%2C2606.29537%2C2606.29705%2C2606.30697%2C2606.31154%2C2606.31270%2C2606.31410%2C2606.31612%2C2607.00502%2C2607.04425&max_results=50"}], "id": "arxiv:2606.16659", "identifiers": {"arxiv": "2606.16659"}, "kind": "paper", "last_verified_at": "2026-10-16T06:33:05Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.16659"}], "name": "FraudSMSWalker: Benchmarking Agentic Large Language Models for SMS-to-Webpage Fraud Detection", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-06-15", "relations": [], "review_status": "curated", "scope": "core", "summary": "FraudSMSWalker tests browser agents on bilingual SMS-to-webpage fraud judgments while hiding URL and reputation shortcuts. Its controlled chains include difficult benign login/payment cases and evaluate whether classifications remain supported by the visible message and page evidence.", "topics": ["safety-security", "evaluation", "robustness"], "updated_at": "2026-06-15", "url": "https://arxiv.org/abs/2606.16659"}
{"aliases": [], "authors": ["Xuelong Dai", "Jianyu Ma", "Boyang Ma", "Biwei Yan", "Yijun Yang", "Yue Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "fea441692131ec677e4a9ae54f9ecef1f31d422e20884b496b124e99392e59c1", "submitted_at": "2026-06-16T15:31:33Z", "updated_at": "2026-10-15T04:29:19Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2606.20717v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.20717", "identifiers": {"arxiv": "2606.20717"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.20717"}], "name": "MIRAGE: Stealthy Visual Prompt Injection for Vulnerability Detection in Web Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-06-16", "relations": [], "review_status": "curated", "scope": "core", "summary": "MIRAGE evaluates targeted next-action hijacking from adversarial images restricted to legitimate third-party webpage regions. It combines diffusion-based image generation with execution-aware optimization and tests the constrained visual-injection setting on web-agent frameworks.", "topics": ["safety-security", "robustness", "evaluation"], "updated_at": "2026-10-15", "url": "https://arxiv.org/abs/2606.20717"}
{"aliases": [], "authors": ["Dayeon Kang", "Hyejun Jeong", "Jade Sheffey", "Pubali Datta", "Amir Houmansadr"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "8b2c6080f0de11c06fe73c30195c7c8529e07de283dc45fa47709072846d385a", "submitted_at": "2026-06-18T20:07:39Z", "updated_at": "2026-06-18T20:07:39Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2606.20910v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.20910", "identifiers": {"arxiv": "2606.20910"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.20910"}], "name": "Whose Agent Are You? Multi-Layer Fingerprinting and Attribution of Autonomous Web Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-06-18", "relations": [], "review_status": "curated", "scope": "core", "summary": "Builds a logging framework that combines network/protocol fingerprints with browser interaction behavior to attribute web-agent traffic. A decision-tree classifier is evaluated across six agent implementations, human browsing and conventional crawlers.", "topics": ["safety-security", "observability", "evaluation"], "updated_at": "2026-06-18", "url": "https://arxiv.org/abs/2606.20910"}
{"aliases": [], "authors": ["Anmol Goel", "Iryna Gurevych"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "da8190bc5afc0bae400fc8ab5e7d80a7f34488c3400dce4477ca61635b19d5b7", "submitted_at": "2026-06-22T11:36:58Z", "updated_at": "2026-09-10T18:49:54Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2606.23189v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.23189", "identifiers": {"arxiv": "2606.23189"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.23189"}], "name": "Capable but Careless: Do Computer-Use Agents Follow Contextual Integrity?", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-06-22", "relations": [], "review_status": "curated", "scope": "core", "summary": "AgentCIBench evaluates whether computer-use agents disclose information in ways that violate its original context. Deterministically scored scenarios test nearby sensitive UI content, ambiguous requests and recipient mismatches, including end-to-end agent execution.", "topics": ["evaluation", "safety-security", "cross-app"], "updated_at": "2026-09-10", "url": "https://arxiv.org/abs/2606.23189"}
{"aliases": [], "authors": ["Aradhana Nayak", "Mussadiq Nazeer", "Wang Peng", "Feng Liu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "3c5a2d0e461b011e2d5b9dbd73012d007261d8df37bd1d687fa3bd86dc6f2246", "submitted_at": "2026-06-24T11:18:01Z", "updated_at": "2026-06-24T11:18:01Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2606.25705v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.25705", "identifiers": {"arxiv": "2606.25705"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.25705"}], "name": "GUI agent: Guided Exploration of User-Sensitive Screens", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-06-24", "relations": [], "review_status": "curated", "scope": "core", "summary": "This short paper develops an explorer agent that expands from a demonstrated GUI task to discover requests likely to reach user-sensitive screens. The proposed exploration supports identifying situations where task execution should be handed back to the user.", "topics": ["safety-security", "human-in-the-loop", "data-generation"], "updated_at": "2026-06-24", "url": "https://arxiv.org/abs/2606.25705"}
{"aliases": [], "authors": ["Divake Kumar", "Sina Tayebati", "Devashri Naik", "Amanda Sofie Rios", "Nilesh Ahuja", "Omesh Tickoo", "Ranganath Krishnan", "Amit Ranjan Trivedi"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "628d08bdbf23ae9e599bf767f0503cecedcde72dfbfcc183f9ea9f044c6f2d44", "submitted_at": "2026-06-24T12:34:28Z", "updated_at": "2026-09-29T20:55:51Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2606.25760v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.25760", "identifiers": {"arxiv": "2606.25760"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.25760"}], "name": "Uncertainty Quantification for Computer-Use Agents: A Benchmark across Vision-Language Models and GUI Grounding Datasets", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-06-24", "relations": [], "review_status": "curated", "scope": "core", "summary": "Argus benchmarks post-hoc uncertainty methods for single-step GUI grounding across open-weight and closed-source observation regimes. It tests calibration, rejection, error severity and conformal click regions, finding that uncertainty rankings transfer less reliably across model/interface regimes than across datasets for one model.", "topics": ["grounding", "evaluation", "robustness", "safety-security"], "updated_at": "2026-09-29", "url": "https://arxiv.org/abs/2606.25760"}
{"aliases": [], "authors": ["Iliana Fayolle", "Sihem Bouhenniche", "Samuel Pélissier", "Pierre Laperdrix", "Clémentine Maurice", "Walter Rudametkin"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "770c901e3490b2a43b02296136825e8e6bd035fb7494ea74b3a9af74be1afed1", "submitted_at": "2026-06-29T10:51:13Z", "updated_at": "2026-06-29T10:51:13Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2606.30119v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:36Z", "role": "primary", "sha256": "aeacc1a41bb90f3d296823356174fe77d1547ffbe7cd56652cf9d98fca09206c", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=200&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2606.30119", "identifiers": {"arxiv": "2606.30119"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:36Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2606.30119"}], "name": "On the Internet, Nobody Knows You're an LLM Bot: Unmasking Web Agents with Multi-Layer Fingerprinting", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-06-29", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates web agents against anti-bot mechanisms on instrumented honeysites and analyzes network, HTTP and browser fingerprints. The study compares resistance to blocking with the ability to distinguish agents, humans and agent implementations.", "topics": ["safety-security", "observability", "evaluation"], "updated_at": "2026-06-29", "url": "https://arxiv.org/abs/2606.30119"}
{"aliases": [], "authors": ["Zidong Zhang", "Zhentao Xie", "Wenrui Diao", "Jianliang Wu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "cd4d31e477d5915626d471682e52e5cd6083d9b0c67196f6de5de2a29c6a3058", "submitted_at": "2026-07-01T02:17:19Z", "updated_at": "2026-07-14T03:01:43Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2607.00333v2"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2607.00333", "identifiers": {"arxiv": "2607.00333"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2607.00333"}], "name": "(A)I Sees What You Don't: Exploiting New Attack Surfaces in Third-Party Mobile Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-07-01", "relations": [], "review_status": "curated", "scope": "core", "summary": "Analyzes cross-app privilege escalation and visual attack surfaces introduced by third-party VLM-powered mobile agents.", "topics": ["safety-security", "mobile-navigation"], "updated_at": "2026-07-14", "url": "https://arxiv.org/abs/2607.00333"}
{"aliases": [], "authors": ["Guijia Zhang", "Yuxun Chen", "Yuheng Qi", "Harry Yang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e7fdae5d6e5ea17639fe0e2b8e2d655df7201a438d84acdeb3e0d02a78051ae2", "submitted_at": "2026-07-05T14:33:29Z", "updated_at": "2026-07-18T16:56:38Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2607.04334v2"}}, "environments": ["cross-platform"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2607.04334", "identifiers": {"arxiv": "2607.04334"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2607.04334"}], "name": "Do GUI Agents Believe Their Eyes? Diagnosing State-Belief Reliance on Pixels versus Structure", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-07-05", "relations": [], "review_status": "curated", "scope": "core", "summary": "Defines visual state reliance and probes whether GUI agents resolve screenshot/structure conflicts using pixels, serialized interface structure or prior beliefs. Paired interventions and live multi-step tests connect misattributed state beliefs to failures and compare consistency-based mitigations.", "topics": ["screen-understanding", "state-tracking", "evaluation", "robustness", "safety-security"], "updated_at": "2026-07-18", "url": "https://arxiv.org/abs/2607.04334"}
{"aliases": [], "authors": ["Woohyuk Choi", "Juhee Kim", "Taehyun Kang", "Jihyeon Jeong", "Luyi Xing", "Byoungyoung Lee"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "c5fb30b8be2c6e452471e6398720291ec59a2c5ef96e64baca0735fb01b7a12e", "submitted_at": "2026-07-06T14:07:49Z", "updated_at": "2026-07-06T14:07:49Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2607.05120v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-17T12:31:32Z", "role": "primary", "sha256": "5128028eea81e3ae80715cb038513012598c0c81a7cbfe8c0d4d6adb6c00e518", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2607.03726%2C2607.05120%2C2607.05465%2C2607.05743%2C2607.06008%2C2607.06595%2C2607.07403%2C2607.07744%2C2607.08252%2C2607.08964%2C2607.10455%2C2607.10608%2C2607.12406%2C2607.12756%2C2607.13085%2C2607.13594%2C2607.13884%2C2607.13988%2C2607.15899%2C2607.16352%2C2607.16636%2C2607.16900%2C2607.17044%2C2607.20064%2C2607.20121%2C2607.21522%2C2607.22738%2C2607.23722%2C2607.24720%2C2607.26300%2C2607.26769%2C2607.27881%2C2607.28229%2C2607.28272%2C2607.28430%2C2608.00548%2C2608.02163%2C2608.03223%2C2608.03451%2C2608.03499%2C2608.05013%2C2608.05430%2C2608.06811%2C2608.07873%2C2608.09268%2C2608.10357%2C2608.10669%2C2608.11274%2C2608.11434%2C2608.15089&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T12:31:32Z", "role": "primary", "sha256": "5128028eea81e3ae80715cb038513012598c0c81a7cbfe8c0d4d6adb6c00e518", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?id_list=2607.03726%2C2607.05120%2C2607.05465%2C2607.05743%2C2607.06008%2C2607.06595%2C2607.07403%2C2607.07744%2C2607.08252%2C2607.08964%2C2607.10455%2C2607.10608%2C2607.12406%2C2607.12756%2C2607.13085%2C2607.13594%2C2607.13884%2C2607.13988%2C2607.15899%2C2607.16352%2C2607.16636%2C2607.16900%2C2607.17044%2C2607.20064%2C2607.20121%2C2607.21522%2C2607.22738%2C2607.23722%2C2607.24720%2C2607.26300%2C2607.26769%2C2607.27881%2C2607.28229%2C2607.28272%2C2607.28430%2C2608.00548%2C2608.02163%2C2608.03223%2C2608.03451%2C2608.03499%2C2608.05013%2C2608.05430%2C2608.06811%2C2608.07873%2C2608.09268%2C2608.10357%2C2608.10669%2C2608.11274%2C2608.11434%2C2608.15089&max_results=50"}], "id": "arxiv:2607.05120", "identifiers": {"arxiv": "2607.05120"}, "kind": "paper", "last_verified_at": "2026-10-17T12:31:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2607.05120"}], "name": "Agent Data Injection Attacks are Realistic Threats to AI Agents", "notes": "Reviewed against the complete retained official abstract. Experimental findings and release availability have not been independently reproduced or verified.", "published_at": "2026-07-06", "relations": [], "review_status": "curated", "scope": "core", "summary": "This study examines agent data injection, where untrusted content impersonates trusted metadata or tool-context data. It reports arbitrary-click vulnerabilities in web agents alongside separate coding-agent attacks, highlighting trust-boundary failures beyond instruction injection.", "topics": ["safety-security", "evaluation", "robustness"], "updated_at": "2026-07-06", "url": "https://arxiv.org/abs/2607.05120"}
{"aliases": [], "authors": ["Kristina Nikolić", "Egor Zverev", "Javier Rando", "Matthew Jagielski", "Edoardo Debenedetti", "Florian Tramèr"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "97d6c625b6ebc08cec7471df543812a91c4252f50d8d35f8d51fbef802dcb5a2", "submitted_at": "2026-07-06T16:22:37Z", "updated_at": "2026-07-06T16:22:37Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2607.05277v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2607.05277", "identifiers": {"arxiv": "2607.05277"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2607.05277"}], "name": "Untrusted Content Masking for Web Agents with Security Guarantees", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-07-06", "relations": [], "review_status": "curated", "scope": "core", "summary": "Untrusted Content Masking uses DOM structure to redact untrusted webpage regions before agent observation. A sandboxed interaction interface separates privileges so browser actions can proceed under the proposed trust-boundary model without exposing protected content to the planner.", "topics": ["safety-security", "sandbox", "web-navigation"], "updated_at": "2026-07-06", "url": "https://arxiv.org/abs/2607.05277"}
{"aliases": [], "authors": ["Yixiong Chen", "Xinyi Bai", "Alan Yuille"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "0473ced7881ed2007aa49c2767860c33719d5bf1c798dfe4771a4061b13d655e", "submitted_at": "2026-07-12T07:05:31Z", "updated_at": "2026-07-12T07:05:31Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2607.10608v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-17T12:31:32Z", "role": "primary", "sha256": "5128028eea81e3ae80715cb038513012598c0c81a7cbfe8c0d4d6adb6c00e518", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2607.03726%2C2607.05120%2C2607.05465%2C2607.05743%2C2607.06008%2C2607.06595%2C2607.07403%2C2607.07744%2C2607.08252%2C2607.08964%2C2607.10455%2C2607.10608%2C2607.12406%2C2607.12756%2C2607.13085%2C2607.13594%2C2607.13884%2C2607.13988%2C2607.15899%2C2607.16352%2C2607.16636%2C2607.16900%2C2607.17044%2C2607.20064%2C2607.20121%2C2607.21522%2C2607.22738%2C2607.23722%2C2607.24720%2C2607.26300%2C2607.26769%2C2607.27881%2C2607.28229%2C2607.28272%2C2607.28430%2C2608.00548%2C2608.02163%2C2608.03223%2C2608.03451%2C2608.03499%2C2608.05013%2C2608.05430%2C2608.06811%2C2608.07873%2C2608.09268%2C2608.10357%2C2608.10669%2C2608.11274%2C2608.11434%2C2608.15089&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T12:31:32Z", "role": "primary", "sha256": "5128028eea81e3ae80715cb038513012598c0c81a7cbfe8c0d4d6adb6c00e518", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?id_list=2607.03726%2C2607.05120%2C2607.05465%2C2607.05743%2C2607.06008%2C2607.06595%2C2607.07403%2C2607.07744%2C2607.08252%2C2607.08964%2C2607.10455%2C2607.10608%2C2607.12406%2C2607.12756%2C2607.13085%2C2607.13594%2C2607.13884%2C2607.13988%2C2607.15899%2C2607.16352%2C2607.16636%2C2607.16900%2C2607.17044%2C2607.20064%2C2607.20121%2C2607.21522%2C2607.22738%2C2607.23722%2C2607.24720%2C2607.26300%2C2607.26769%2C2607.27881%2C2607.28229%2C2607.28272%2C2607.28430%2C2608.00548%2C2608.02163%2C2608.03223%2C2608.03451%2C2608.03499%2C2608.05013%2C2608.05430%2C2608.06811%2C2608.07873%2C2608.09268%2C2608.10357%2C2608.10669%2C2608.11274%2C2608.11434%2C2608.15089&max_results=50"}], "id": "arxiv:2607.10608", "identifiers": {"arxiv": "2607.10608"}, "kind": "paper", "last_verified_at": "2026-10-17T12:31:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2607.10608"}], "name": "The Compliance Trap: Diagnosing How AI Agents Consume Conflicting Memory", "notes": "Reviewed against the complete retained official abstract. Experimental findings and release availability have not been independently reproduced or verified.", "published_at": "2026-07-12", "relations": [], "review_status": "curated", "scope": "core", "summary": "The Entry–Propagation–Recovery framework traces where conflicting memory first changes an agent action, how the effect propagates and whether recovery occurs. WebArena and the controlled MemTrapBench separate memory-consumption failures across these stages.", "topics": ["memory", "safety-security", "robustness", "evaluation", "long-horizon"], "updated_at": "2026-07-12", "url": "https://arxiv.org/abs/2607.10608"}
{"aliases": [], "authors": ["Xue Yu", "Bo Yuan", "Kailin Zhao", "Pengshuai Yang", "Hong Hu", "Junlan Feng"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "6a15963a7fb4dfd488f0e79732d9ea0c9a03d43c1e3d5e50dd1ee8e00f1072be", "submitted_at": "2026-07-17T01:45:54Z", "updated_at": "2026-09-02T01:49:21Z", "version": 3, "versioned_url": "https://arxiv.org/abs/2607.15550v3"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2607.15550", "identifiers": {"arxiv": "2607.15550"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2607.15550"}], "name": "SeerGuard: A Safety Framework for Mobile GUI Agents via World Model Prediction", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-07-17", "relations": [], "review_status": "curated", "scope": "core", "summary": "Uses instruction screening and a safety-augmented world model to assess likely consequences of mobile GUI actions before execution.", "topics": ["safety-security", "world-models"], "updated_at": "2026-09-02", "url": "https://arxiv.org/abs/2607.15550"}
{"aliases": [], "authors": ["Behzad Ousat", "Nikita Turkmen", "Lalchandra Rampersaud", "Dillan Bailey", "Amin Kharraz"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e382b15a9d25e99aa187a72c7d2a33bbeca82b2b347ba580ca0a1d73a8e01d19", "submitted_at": "2026-07-21T03:02:57Z", "updated_at": "2026-07-21T03:02:57Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2607.18659v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2607.18659", "identifiers": {"arxiv": "2607.18659"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2607.18659"}], "name": "Broken Gates: Re-evaluating Web Bot Defenses in the Age of LLM Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-07-21", "relations": [], "review_status": "curated", "scope": "core", "summary": "Measures how autonomous LLM browser agents interact with commercial bot-management and anti-automation defenses on the web.", "topics": ["safety-security", "web-navigation", "evaluation"], "updated_at": "2026-07-21", "url": "https://arxiv.org/abs/2607.18659"}
{"aliases": [], "authors": ["Yashaswi Malla", "Sandra Siby"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "60c93c90e6d4026452810a866a0c49d57afdec15ad5ced30606c925a8e83c8f7", "submitted_at": "2026-07-31T18:27:12Z", "updated_at": "2026-07-31T18:27:12Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2608.00202v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2608.00202", "identifiers": {"arxiv": "2608.00202"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.00202"}], "name": "From Monoliths to Swarms: A Study of Attack Surface Evolution in the Transition to Multi-Agent Web Systems", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-07-31", "relations": [], "review_status": "curated", "scope": "core", "summary": "Categorizes structural attack surfaces that emerge when single-agent browser systems are decomposed into multi-agent web workflows.", "topics": ["safety-security", "web-navigation", "planning"], "updated_at": "2026-07-31", "url": "https://arxiv.org/abs/2608.00202"}
{"aliases": [], "authors": ["Yu Ran", "Wentao Zhao"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "152e21cd4f1f290aafadfdcfaa10e15a03378404540a46b5f1a5464a2d432d23", "submitted_at": "2026-08-04T14:34:41Z", "updated_at": "2026-09-28T13:07:55Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2608.03740v2"}}, "environments": ["cross-platform"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2608.03740", "identifiers": {"arxiv": "2608.03740"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.03740"}], "name": "MissClick: Execution-Aware Adversarial Attacks on Coordinate Generation in GUI Grounding Models", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-08-04", "relations": [], "review_status": "curated", "scope": "core", "summary": "MissClick designs white-box attacks around the numerical effect of coordinate digits on executable GUI clicks. Separate untargeted displacement and targeted place-weighted objectives are evaluated across desktop, web and mobile grounding settings.", "topics": ["safety-security", "grounding", "robustness", "evaluation"], "updated_at": "2026-09-28", "url": "https://arxiv.org/abs/2608.03740"}
{"aliases": [], "authors": ["Longtao Guo", "Zelin Zhang", "Kaifeng Huang", "Yang Shi"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "30e5ce3d86881041491ae507839b2f3d302691aea1d2fdfc92cfc00f35efefe7", "submitted_at": "2026-08-05T12:08:16Z", "updated_at": "2026-08-05T12:08:16Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2608.04741v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2608.04741", "identifiers": {"arxiv": "2608.04741"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.04741"}], "name": "LoginTrap: Uncovering Task-Agnostic Phishing-Style Indirect Prompt Injection Attacks against LLM-based Web Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-08-05", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces LoginTrap, a black-box indirect-injection evaluation in which attacker-controlled page context and login flows induce unauthorized authentication. It measures end-to-end private-data leakage across web-agent backbones, architectures and defenses.", "topics": ["safety-security", "robustness", "evaluation"], "updated_at": "2026-08-05", "url": "https://arxiv.org/abs/2608.04741"}
{"aliases": [], "authors": ["Dongsheng Chen", "Yuxuan Li", "Guanhua Chen", "Jiaxin Zhang", "Xiangyu Zhao", "Lei Ma", "Xin Yao", "Xuetao Wei"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "5f8b84135f72558eea1fae428df7cb30f62232631de49e690e73010aeca22d4c", "submitted_at": "2026-08-05T12:23:16Z", "updated_at": "2026-08-05T12:23:16Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2608.04755v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2608.04755", "identifiers": {"arxiv": "2608.04755"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.04755"}], "name": "\"Allow\" to Achieve, Over-Privileged Inadvertently: The Unintended Cost of Task-Completion-Driven Pop-up Decisions in Mobile GUI Agents", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2026-08-05", "relations": [], "review_status": "curated", "scope": "core", "summary": "This permission-literacy study tests whether mobile GUI agents grant only task-necessary permissions. Controlled changes to task context and requester identity expose authorization biases and the trade-off between suppressing unnecessary and legitimate grants.", "topics": ["safety-security", "evaluation", "robustness"], "updated_at": "2026-08-05", "url": "https://arxiv.org/abs/2608.04755"}
{"aliases": [], "authors": ["Rahul Deivasigamani", "Sayeda Faatin Alvi", "Derqui Andrea", "Kaushal Punjabi", "Stjepan Picek"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "64e4845ce85dd67b7790cf5de357f12deb72b71eb72031298e1064809099f063", "submitted_at": "2026-08-09T22:13:55Z", "updated_at": "2026-08-09T22:13:55Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2608.08939v1"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2608.08939", "identifiers": {"arxiv": "2608.08939"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.08939"}], "name": "Not an A11y: How Android Accessibility Exposes Mobile AI Agents to Indirect Prompt Injection", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-08-09", "relations": [], "review_status": "curated", "scope": "core", "summary": "Demonstrates that unsanitized Android accessibility tree metadata exposes mobile GUI agents to stealthy indirect prompt injection attacks.", "topics": ["safety-security", "mobile-navigation", "screen-understanding"], "updated_at": "2026-08-09", "url": "https://arxiv.org/abs/2608.08939"}
{"aliases": [], "authors": ["Thilo Reintjes", "Sivajeet Chand", "Derui Zhu", "Sushant Kumar Pandey", "Alexander Pretschner"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "f8a6bd69d421bdfb9a21c22e715a59d0afb0b0cab75510f28ff569903ecffae5", "submitted_at": "2026-08-14T09:36:33Z", "updated_at": "2026-08-14T09:36:33Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2608.14131v1"}}, "environments": ["desktop", "windows"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2608.14131", "identifiers": {"arxiv": "2608.14131"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.14131"}], "name": "LegacyWorld: Atomicity-Aware Evaluation of GUI Agents for Legacy Workflows", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-08-14", "relations": [], "review_status": "curated", "scope": "core", "summary": "LegacyWorld evaluates Windows GUI workflows under an atomicity criterion: finish correctly or fail without persistent unintended changes. It compares expert prompts with prompts derived from recorded golden paths and separates useful completion, safe failure and non-atomic side effects.", "topics": ["evaluation", "safety-security", "workflow", "recording"], "updated_at": "2026-08-14", "url": "https://arxiv.org/abs/2608.14131"}
{"aliases": [], "authors": ["Sujin Chen", "Lijun Li", "Tianyi Du", "Jing Shao"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "c49a67b38314152b83e793b9ade7ce4102a44aa27c331504a8758b94df906c67", "submitted_at": "2026-08-18T11:33:22Z", "updated_at": "2026-08-18T11:33:22Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2608.17659v1"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2608.17659", "identifiers": {"arxiv": "2608.17659"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.17659"}], "name": "MobileWorldSafety: Benchmarking GUI Agent Safety Against Environmental Injection Attacks in Android Apps", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-08-18", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates environmental injection risks in Android tasks using programmatic risk checks and model-based adjudication of ambiguous cases.", "topics": ["safety-security", "evaluation"], "updated_at": "2026-08-18", "url": "https://arxiv.org/abs/2608.17659"}
{"aliases": [], "authors": ["Lin-Fa Lee", "YI-YU Chang", "Kuo-Hui Yeh"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "1bddfe5b5b2312e9b71d01dbb6bb23a8b0ad1c8cd787e4e98449b71a2ac2d678", "submitted_at": "2026-08-25T03:16:45Z", "updated_at": "2026-08-25T03:16:45Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2608.24017v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:33:13Z", "role": "primary", "sha256": "ca895b2f0dcd74816fb5e775a7161a0932ff7120a72477791f5e4c0bb1ead88e", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2608.17319%2C2608.18307%2C2608.20319%2C2608.20707%2C2608.20797%2C2608.21794%2C2608.21830%2C2608.21832%2C2608.21898%2C2608.22077%2C2608.22533%2C2608.22577%2C2608.22847%2C2608.24015%2C2608.24017%2C2608.24099%2C2608.24174%2C2608.24680%2C2608.24848%2C2608.25777%2C2608.26200%2C2608.26204%2C2608.27477%2C2608.27508%2C2608.27808%2C2608.27866%2C2608.29305%2C2608.30038%2C2608.30207%2C2609.00028%2C2609.00048%2C2609.00524%2C2609.02057%2C2609.02149%2C2609.02154%2C2609.02309%2C2609.02885%2C2609.03438%2C2609.04850%2C2609.04869%2C2609.05374%2C2609.05511%2C2609.05513%2C2609.05535%2C2609.06530%2C2609.06792%2C2609.07603%2C2609.07712%2C2609.09212%2C2609.10031&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:33:13Z", "role": "primary", "sha256": "ca895b2f0dcd74816fb5e775a7161a0932ff7120a72477791f5e4c0bb1ead88e", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2608.17319%2C2608.18307%2C2608.20319%2C2608.20707%2C2608.20797%2C2608.21794%2C2608.21830%2C2608.21832%2C2608.21898%2C2608.22077%2C2608.22533%2C2608.22577%2C2608.22847%2C2608.24015%2C2608.24017%2C2608.24099%2C2608.24174%2C2608.24680%2C2608.24848%2C2608.25777%2C2608.26200%2C2608.26204%2C2608.27477%2C2608.27508%2C2608.27808%2C2608.27866%2C2608.29305%2C2608.30038%2C2608.30207%2C2609.00028%2C2609.00048%2C2609.00524%2C2609.02057%2C2609.02149%2C2609.02154%2C2609.02309%2C2609.02885%2C2609.03438%2C2609.04850%2C2609.04869%2C2609.05374%2C2609.05511%2C2609.05513%2C2609.05535%2C2609.06530%2C2609.06792%2C2609.07603%2C2609.07712%2C2609.09212%2C2609.10031&max_results=50"}], "id": "arxiv:2608.24017", "identifiers": {"arxiv": "2608.24017"}, "kind": "paper", "last_verified_at": "2026-10-16T06:33:13Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.24017"}], "name": "WebMCP-Phalanx: Enforcing and Characterizing Trust Boundaries for Browser-Integrated LLM Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-08-25", "relations": [], "review_status": "curated", "scope": "adjacent", "summary": "Proposes WebMCP-Phalanx, a browser-integrated runtime that binds page-exposed tools to their registering principals and separates content inspection from privileged invocation. It studies lifecycle and prompt-injection attacks, including a timing bypass that motivates validating all visible metadata before tool use.", "topics": ["safety-security", "tool-integration"], "updated_at": "2026-08-25", "url": "https://arxiv.org/abs/2608.24017"}
{"aliases": [], "authors": ["Lily Bundgaard Wanscher", "Markus Heidemann Lorensen", "Mohammed Ammad Shafiq", "Mahyar Tourchi Moghaddam", "Mina Alipour"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "046d93a43dd5a011809a780c7d998f98ee0abb3e931d733b8ad7807b86050ec7", "submitted_at": "2026-07-26T17:21:24Z", "updated_at": "2026-07-26T17:21:24Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2608.24913v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}], "id": "arxiv:2608.24913", "identifiers": {"arxiv": "2608.24913"}, "kind": "paper", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.24913"}], "name": "From Blind Edits to Verified Repair: Building Trustworthy User-Side LLM Agents for Web Accessibility", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-07-26", "relations": [], "review_status": "curated", "scope": "core", "summary": "Builds a local browser-extension agent that generates and verifies additive CSS repairs for WCAG web accessibility violations.", "topics": ["ui-design", "web-navigation", "safety-security"], "updated_at": "2026-07-26", "url": "https://arxiv.org/abs/2608.24913"}
{"aliases": [], "authors": ["Joy Chen", "Alejandro Castillejo Munoz", "Pierluca D'Oro", "Yuxuan Sun", "Chloe Evans", "Joseph Tighe"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "50c26a7ab741ecb3cb855305ee4f7d5b5729296248d650c469600aeb73101781", "submitted_at": "2026-08-25T23:21:57Z", "updated_at": "2026-09-15T04:43:00Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2608.26204v2"}}, "environments": ["desktop", "cross-platform"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2608.26204", "identifiers": {"arxiv": "2608.26204"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.26204"}], "name": "ADeptS-Bench: Measuring the Trustworthiness of Computer Use Agents Across Devices", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-08-25", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates interface-embedded threats and whether computer-use agents seek clarification when mobile or desktop instructions are ambiguous.", "topics": ["safety-security", "evaluation", "human-in-the-loop"], "updated_at": "2026-09-15", "url": "https://arxiv.org/abs/2608.26204"}
{"aliases": [], "authors": ["Divake Kumar", "Sina Tayebati", "Devashri Naik", "Amanda Sofie Rios", "Nilesh Ahuja", "Omesh Tickoo", "Ranganath Krishnan", "Amit Ranjan Trivedi"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "14864aae8f977d474ff3e7dd3e38951374862abfb8077395f2dc5a9e1cc5a489", "submitted_at": "2026-08-28T00:50:47Z", "updated_at": "2026-08-28T00:50:47Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2608.27808v1"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2608.27808", "identifiers": {"arxiv": "2608.27808"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.27808"}], "name": "CURA: Certified Runtime Alarms for Computer-Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-08-28", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces CURA, a telemetry-based runtime monitor that turns computer-use trajectories into sequential failure alarms and can trigger mid-task oversight. Its certificate controls false alarms, while the study separately reports detection performance and limits of retrospective monitoring.", "topics": ["observability", "safety-security", "evaluation"], "updated_at": "2026-08-28", "url": "https://arxiv.org/abs/2608.27808"}
{"aliases": [], "authors": ["Sanket Badhe", "Deep Shah", "Priyanka Tiwari", "Nehal Kathrotia"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "13b6833243a543eb156ce7f4d943f8a3c608b3b38b7eaa77605ca7484cdedc43", "submitted_at": "2026-08-30T06:36:42Z", "updated_at": "2026-09-28T17:25:01Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2608.29596v2"}}, "environments": ["cross-platform"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:54:12Z", "role": "primary", "sha256": "5f5bae19420f79231516d4bcf604ea18d4f930fc7790cde5a3e7fcb8d41a1126", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2508.19461%2C2508.19493%2C2509.21782%2C2509.25140%2C2510.03485%2C2510.07809%2C2510.14276%2C2510.27140%2C2510.27210%2C2511.13248%2C2512.08417%2C2601.07263%2C2602.13516%2C2603.00476%2C2603.11862%2C2603.18762%2C2603.23791%2C2604.00986%2C2604.16515%2C2604.16548%2C2604.24348%2C2605.05509%2C2605.08876%2C2605.20833%2C2605.24785%2C2607.00333%2C2607.18659%2C2607.25255%2C2608.00202%2C2608.08939%2C2608.24913%2C2608.29596&max_results=32"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T13:00:09.204016+00:00", "role": "primary", "sha256": "d112d0a83472b87de538d0290efdd2d0dd0d21e878acca5f383da5baa880fe88", "supports": ["scope", "environments", "notes"], "url": "https://arxiv.org/html/2608.29596v2"}], "id": "arxiv:2608.29596", "identifiers": {"arxiv": "2608.29596"}, "kind": "survey", "last_verified_at": "2026-10-18T10:54:12Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.29596"}], "name": "A Systematic Survey of Agentic Skills: Architecture, Lifecycle, and Security", "notes": "Adjacent broad survey of procedural agent skills. Full-text Section 4.4.2 explicitly covers desktop, mobile, and web interface skills; surveyed systems and their reported results are not independently reproduced.", "published_at": "2026-08-30", "relations": [], "review_status": "curated", "scope": "adjacent", "summary": "Surveys modular agentic skill architectures, skill lifecycles, and security risks in autonomous tool- and interface-using agents.", "topics": ["survey", "workflow", "memory", "safety-security"], "updated_at": "2026-09-28", "url": "https://arxiv.org/abs/2608.29596"}
{"aliases": [], "authors": ["Mingshuo Wang", "Hanqing Guo", "Huining Li", "Yuliang Fu", "Jing Xu", "Chenhan Xu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "7b17f858c9d6992f6a8d8f21cc79d468b0f1c81fc44084045791d9c3acacd6d0", "submitted_at": "2026-08-30T20:55:21Z", "updated_at": "2026-08-30T20:55:21Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2608.30038v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:33:13Z", "role": "primary", "sha256": "ca895b2f0dcd74816fb5e775a7161a0932ff7120a72477791f5e4c0bb1ead88e", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2608.17319%2C2608.18307%2C2608.20319%2C2608.20707%2C2608.20797%2C2608.21794%2C2608.21830%2C2608.21832%2C2608.21898%2C2608.22077%2C2608.22533%2C2608.22577%2C2608.22847%2C2608.24015%2C2608.24017%2C2608.24099%2C2608.24174%2C2608.24680%2C2608.24848%2C2608.25777%2C2608.26200%2C2608.26204%2C2608.27477%2C2608.27508%2C2608.27808%2C2608.27866%2C2608.29305%2C2608.30038%2C2608.30207%2C2609.00028%2C2609.00048%2C2609.00524%2C2609.02057%2C2609.02149%2C2609.02154%2C2609.02309%2C2609.02885%2C2609.03438%2C2609.04850%2C2609.04869%2C2609.05374%2C2609.05511%2C2609.05513%2C2609.05535%2C2609.06530%2C2609.06792%2C2609.07603%2C2609.07712%2C2609.09212%2C2609.10031&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:33:13Z", "role": "primary", "sha256": "ca895b2f0dcd74816fb5e775a7161a0932ff7120a72477791f5e4c0bb1ead88e", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2608.17319%2C2608.18307%2C2608.20319%2C2608.20707%2C2608.20797%2C2608.21794%2C2608.21830%2C2608.21832%2C2608.21898%2C2608.22077%2C2608.22533%2C2608.22577%2C2608.22847%2C2608.24015%2C2608.24017%2C2608.24099%2C2608.24174%2C2608.24680%2C2608.24848%2C2608.25777%2C2608.26200%2C2608.26204%2C2608.27477%2C2608.27508%2C2608.27808%2C2608.27866%2C2608.29305%2C2608.30038%2C2608.30207%2C2609.00028%2C2609.00048%2C2609.00524%2C2609.02057%2C2609.02149%2C2609.02154%2C2609.02309%2C2609.02885%2C2609.03438%2C2609.04850%2C2609.04869%2C2609.05374%2C2609.05511%2C2609.05513%2C2609.05535%2C2609.06530%2C2609.06792%2C2609.07603%2C2609.07712%2C2609.09212%2C2609.10031&max_results=50"}], "id": "arxiv:2608.30038", "identifiers": {"arxiv": "2608.30038"}, "kind": "paper", "last_verified_at": "2026-10-16T06:33:13Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.30038"}], "name": "ActReal: System-Level Mobile Agents Challenge Mobile Automation Detection", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-08-30", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces ActReal, which translates semantic mobile-agent actions into coordinated touch and sensor signals to study automation-detection weaknesses. It evaluates detectors that observe touch events alone or together with inertial measurements under a privileged executor threat model.", "topics": ["safety-security", "mobile-navigation", "evaluation"], "updated_at": "2026-08-30", "url": "https://arxiv.org/abs/2608.30038"}
{"aliases": [], "authors": ["Chen Xiong", "Zhiyuan He", "Pin-Yu Chen", "Stjepan Picek", "Tsung-Yi Ho"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "87c3f58eebe2db437b1166ff8e83b336db4cde37684a7cd86f61c4b34fca7237", "submitted_at": "2026-08-31T03:39:43Z", "updated_at": "2026-10-03T10:41:41Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2608.30207v2"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2608.30207", "identifiers": {"arxiv": "2608.30207"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2608.30207"}], "name": "SIR: Self-improving Red-teaming for Compute Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-08-31", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces SIR, which iteratively refines a reusable library of prompt-injection strategies from failed computer-use attack trajectories. Its evaluation uses deterministic system-state checks and counts attacks only when the adversarial objective and legitimate task both complete.", "topics": ["safety-security", "evaluation", "data-generation"], "updated_at": "2026-10-03", "url": "https://arxiv.org/abs/2608.30207"}
{"aliases": [], "authors": ["Venus Team", "Zhuohan Cai", "Haoxing Chen", "Jiaxuan Chen", "Weizhi Chen", "Changlong Gao", "Zhangxuan Gu", "Yuan Guo", "Yusong Hu", "Jianrong Jiang", "Jianguo Li", "Runze Li", "Jinzhen Lin", "Zhenyu Ma", "Changhua Meng", "Han Peng", "Xinyu Qiu", "Shuheng Shen", "Zhongyi Shui", "Weiqiang Wang", "Ming Wen", "Zhuoer Xu", "Hang Yan", "Kaiwen Yang", "Ruilin Yao", "Nanjun Yu", "Zhengwen Zeng", "Lianrui Zhang", "Yunzhu Zhang", "Zhe Zhao", "Beitong Zhou"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "34383fa6c6e5cca059c53fd0452d93eec9ac6a96271ddc14f81ddb96fdeaceec", "submitted_at": "2026-08-27T17:15:33Z", "updated_at": "2026-08-27T17:15:33Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.00028v1"}}, "environments": ["web", "desktop", "cross-platform"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:33:13Z", "role": "primary", "sha256": "ca895b2f0dcd74816fb5e775a7161a0932ff7120a72477791f5e4c0bb1ead88e", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2608.17319%2C2608.18307%2C2608.20319%2C2608.20707%2C2608.20797%2C2608.21794%2C2608.21830%2C2608.21832%2C2608.21898%2C2608.22077%2C2608.22533%2C2608.22577%2C2608.22847%2C2608.24015%2C2608.24017%2C2608.24099%2C2608.24174%2C2608.24680%2C2608.24848%2C2608.25777%2C2608.26200%2C2608.26204%2C2608.27477%2C2608.27508%2C2608.27808%2C2608.27866%2C2608.29305%2C2608.30038%2C2608.30207%2C2609.00028%2C2609.00048%2C2609.00524%2C2609.02057%2C2609.02149%2C2609.02154%2C2609.02309%2C2609.02885%2C2609.03438%2C2609.04850%2C2609.04869%2C2609.05374%2C2609.05511%2C2609.05513%2C2609.05535%2C2609.06530%2C2609.06792%2C2609.07603%2C2609.07712%2C2609.09212%2C2609.10031&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:33:13Z", "role": "primary", "sha256": "ca895b2f0dcd74816fb5e775a7161a0932ff7120a72477791f5e4c0bb1ead88e", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2608.17319%2C2608.18307%2C2608.20319%2C2608.20707%2C2608.20797%2C2608.21794%2C2608.21830%2C2608.21832%2C2608.21898%2C2608.22077%2C2608.22533%2C2608.22577%2C2608.22847%2C2608.24015%2C2608.24017%2C2608.24099%2C2608.24174%2C2608.24680%2C2608.24848%2C2608.25777%2C2608.26200%2C2608.26204%2C2608.27477%2C2608.27508%2C2608.27808%2C2608.27866%2C2608.29305%2C2608.30038%2C2608.30207%2C2609.00028%2C2609.00048%2C2609.00524%2C2609.02057%2C2609.02149%2C2609.02154%2C2609.02309%2C2609.02885%2C2609.03438%2C2609.04850%2C2609.04869%2C2609.05374%2C2609.05511%2C2609.05513%2C2609.05535%2C2609.06530%2C2609.06792%2C2609.07603%2C2609.07712%2C2609.09212%2C2609.10031&max_results=50"}], "id": "arxiv:2609.00028", "identifiers": {"arxiv": "2609.00028"}, "kind": "paper", "last_verified_at": "2026-10-16T06:33:13Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.00028"}], "name": "UI-Venus-2 Technical Report", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-08-27", "relations": [], "review_status": "curated", "scope": "core", "summary": "Describes a GUI foundation agent trained through expanded environments, function-grounded tasks, and trajectory verification with safety-aware execution mechanisms.", "topics": ["automation", "reinforcement-learning", "safety-security"], "updated_at": "2026-08-27", "url": "https://arxiv.org/abs/2609.00028"}
{"aliases": [], "authors": ["Yimeng Liu", "Hua Huang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "b462dddd45e6b04a3bf6fa0bc73576a23a5f9617f61aeb257a845c0f456f731d", "submitted_at": "2026-09-02T06:06:43Z", "updated_at": "2026-09-02T06:06:43Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.02154v1"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:33:13Z", "role": "primary", "sha256": "ca895b2f0dcd74816fb5e775a7161a0932ff7120a72477791f5e4c0bb1ead88e", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2608.17319%2C2608.18307%2C2608.20319%2C2608.20707%2C2608.20797%2C2608.21794%2C2608.21830%2C2608.21832%2C2608.21898%2C2608.22077%2C2608.22533%2C2608.22577%2C2608.22847%2C2608.24015%2C2608.24017%2C2608.24099%2C2608.24174%2C2608.24680%2C2608.24848%2C2608.25777%2C2608.26200%2C2608.26204%2C2608.27477%2C2608.27508%2C2608.27808%2C2608.27866%2C2608.29305%2C2608.30038%2C2608.30207%2C2609.00028%2C2609.00048%2C2609.00524%2C2609.02057%2C2609.02149%2C2609.02154%2C2609.02309%2C2609.02885%2C2609.03438%2C2609.04850%2C2609.04869%2C2609.05374%2C2609.05511%2C2609.05513%2C2609.05535%2C2609.06530%2C2609.06792%2C2609.07603%2C2609.07712%2C2609.09212%2C2609.10031&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:33:13Z", "role": "primary", "sha256": "ca895b2f0dcd74816fb5e775a7161a0932ff7120a72477791f5e4c0bb1ead88e", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2608.17319%2C2608.18307%2C2608.20319%2C2608.20707%2C2608.20797%2C2608.21794%2C2608.21830%2C2608.21832%2C2608.21898%2C2608.22077%2C2608.22533%2C2608.22577%2C2608.22847%2C2608.24015%2C2608.24017%2C2608.24099%2C2608.24174%2C2608.24680%2C2608.24848%2C2608.25777%2C2608.26200%2C2608.26204%2C2608.27477%2C2608.27508%2C2608.27808%2C2608.27866%2C2608.29305%2C2608.30038%2C2608.30207%2C2609.00028%2C2609.00048%2C2609.00524%2C2609.02057%2C2609.02149%2C2609.02154%2C2609.02309%2C2609.02885%2C2609.03438%2C2609.04850%2C2609.04869%2C2609.05374%2C2609.05511%2C2609.05513%2C2609.05535%2C2609.06530%2C2609.06792%2C2609.07603%2C2609.07712%2C2609.09212%2C2609.10031&max_results=50"}], "id": "arxiv:2609.02154", "identifiers": {"arxiv": "2609.02154"}, "kind": "paper", "last_verified_at": "2026-10-16T06:33:13Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.02154"}], "name": "WiP: Characterizing and Defending Against Mobile-Agent-Driven MFA Automation", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-09-02", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies how mobile automation can combine authentication factors within one agent environment and evaluates motion sensing as a user-presence signal.", "topics": ["safety-security", "evaluation"], "updated_at": "2026-09-02", "url": "https://arxiv.org/abs/2609.02154"}
{"aliases": [], "authors": ["Zhaoyuan Huang", "Tianjie Ju", "Pengzhou Cheng", "Zheng Wu", "Yansi Li", "Chuanbiao Song", "Jun Lan", "Huijia Zhu", "Weiqiang Wang", "Zhuosheng Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "a59f7d54845de6efebf6365a188e61e41fdf31e9ebe894b1d01ce7b86d6be4ec", "submitted_at": "2026-09-03T06:48:41Z", "updated_at": "2026-09-03T06:48:41Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.03438v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.03438", "identifiers": {"arxiv": "2609.03438"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.03438"}], "name": "Do GUI Agents Know When Not to Act? Enabling Conflict-Aware Termination for Multimodal GUI Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-09-03", "relations": [], "review_status": "curated", "scope": "core", "summary": "CONFLICTGUI evaluates termination under instruction conflicts and mismatches between instructions and interface context. CONFLICTGUARD couples feasibility checks with conditional action modulation to reduce unnecessary execution while tracking normal-task behavior.", "topics": ["safety-security", "evaluation", "robustness", "planning"], "updated_at": "2026-09-03", "url": "https://arxiv.org/abs/2609.03438"}
{"aliases": [], "authors": ["Zhihao Liu", "Hongyu Sun", "Zhiyuan Fu", "Xiaonan Duan", "Jice Wang", "Shangru Zhao", "Weizhi Meng", "Wuxin Yang", "Yangfan Zhou", "Yuqing Zhang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "5a3a7a81ffe43885d9c8362df4c8aff08000304fe60477f484a732caf35124b2", "submitted_at": "2026-09-06T03:49:45Z", "updated_at": "2026-09-06T03:49:45Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.09212v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.09212", "identifiers": {"arxiv": "2609.09212"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.09212"}], "name": "AgentHijack: Visual Patch Attacks on Multimodal Computer-Use Agents", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2026-09-06", "relations": [], "review_status": "curated", "scope": "core", "summary": "AgentHijack evaluates whether visual patch effects propagate from screenshots through model outputs and action parsing into real environment changes. Its controlled web-page experiments distinguish intermediate attack signals from verified end-to-end consequences.", "topics": ["safety-security", "evaluation", "robustness"], "updated_at": "2026-09-06", "url": "https://arxiv.org/abs/2609.09212"}
{"aliases": [], "authors": ["Qihang Cen", "Tianshuo Cong", "Da Song", "Xinlei He", "Jiaxing Song", "Ke Xu", "Qi Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "1eda33b358d4c4caf8c1fb0b1bd92698c3b37cee995d820f40ecffe82010edb4", "submitted_at": "2026-09-12T10:55:54Z", "updated_at": "2026-09-12T10:55:54Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.13873v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.13873", "identifiers": {"arxiv": "2609.13873"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.13873"}], "name": "PriMobiBench: Characterizing Visual Privacy Leakage in VLM-Driven Mobile GUI Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-09-12", "relations": [], "review_status": "curated", "scope": "core", "summary": "PriMobiBench evaluates direct privacy leakage and user profiling from mobile GUI screenshot trajectories, supported by the MobiLeak trace dataset. It also tests masking sensitive, task-irrelevant interface elements before cloud processing and measures the resulting privacy and utility trade-off.", "topics": ["evaluation", "safety-security", "screen-understanding", "mobile-navigation"], "updated_at": "2026-09-12", "url": "https://arxiv.org/abs/2609.13873"}
{"aliases": [], "authors": ["Heng Li", "Fulin Zhao", "Zhe Geng", "Zhiyuan Yao", "Wei Yuan", "Xiapu Luo"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "b86516b7695d23ffc9179c0829209dc28b96c3c819525eab62a293a68ab5daf0", "submitted_at": "2026-09-15T07:02:10Z", "updated_at": "2026-09-15T07:02:10Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.16732v1"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:33:17Z", "role": "primary", "sha256": "5d7c9ff24833b254d33a41a7c4247e830e4d9c371881a1e16151bad5c71cc8b3", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2609.10297%2C2609.10451%2C2609.12394%2C2609.12404%2C2609.13287%2C2609.13491%2C2609.13548%2C2609.13873%2C2609.14239%2C2609.15134%2C2609.15307%2C2609.15364%2C2609.15457%2C2609.16251%2C2609.16635%2C2609.16732%2C2609.16900%2C2609.17653%2C2609.17885%2C2609.18411%2C2609.18690%2C2609.19125%2C2609.19523%2C2609.19843%2C2609.22083%2C2609.22724%2C2609.24277%2C2609.24890%2C2609.25769%2C2609.27273%2C2609.27307%2C2609.27353%2C2609.27461%2C2609.29892%2C2609.30186%2C2609.30604%2C2609.30887%2C2609.31957%2C2609.32036%2C2609.32130%2C2609.32166%2C2609.32351%2C2609.32385%2C2609.32600%2C2609.32679%2C2609.32750%2C2609.32915%2C2609.33304%2C2609.33646%2C2609.34113&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:33:17Z", "role": "primary", "sha256": "5d7c9ff24833b254d33a41a7c4247e830e4d9c371881a1e16151bad5c71cc8b3", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?id_list=2609.10297%2C2609.10451%2C2609.12394%2C2609.12404%2C2609.13287%2C2609.13491%2C2609.13548%2C2609.13873%2C2609.14239%2C2609.15134%2C2609.15307%2C2609.15364%2C2609.15457%2C2609.16251%2C2609.16635%2C2609.16732%2C2609.16900%2C2609.17653%2C2609.17885%2C2609.18411%2C2609.18690%2C2609.19125%2C2609.19523%2C2609.19843%2C2609.22083%2C2609.22724%2C2609.24277%2C2609.24890%2C2609.25769%2C2609.27273%2C2609.27307%2C2609.27353%2C2609.27461%2C2609.29892%2C2609.30186%2C2609.30604%2C2609.30887%2C2609.31957%2C2609.32036%2C2609.32130%2C2609.32166%2C2609.32351%2C2609.32385%2C2609.32600%2C2609.32679%2C2609.32750%2C2609.32915%2C2609.33304%2C2609.33646%2C2609.34113&max_results=50"}], "id": "arxiv:2609.16732", "identifiers": {"arxiv": "2609.16732"}, "kind": "paper", "last_verified_at": "2026-10-16T06:33:17Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.16732"}], "name": "When Agents See Differently: Exposing UI Desynchronization Threats in Mobile Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-09-15", "relations": [], "review_status": "curated", "scope": "core", "summary": "Investigates mismatches between what humans see on a mobile display and what agents receive through screenshots or accessibility metadata. An automated evaluation framework embeds instruction-agnostic perturbations into functional APK clones and compares static and dynamic agent deviations with human noticeability.", "topics": ["safety-security", "screen-understanding", "evaluation", "robustness"], "updated_at": "2026-09-15", "url": "https://arxiv.org/abs/2609.16732"}
{"aliases": [], "authors": ["Haya Halimeh", "Sascha Kaltenpoth", "Kevin Bösch", "Oliver Müller"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "70a7588abcf345418ed759fb2bdb96fc60c6e1444a3e0dc98766510c755f8478", "submitted_at": "2026-09-17T07:54:15Z", "updated_at": "2026-09-17T07:54:15Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.19843v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.19843", "identifiers": {"arxiv": "2609.19843"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.19843"}], "name": "A Dual-Process Perspective on Nudge Susceptibility in LLM-Based GUI Agents", "notes": "The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced.", "published_at": "2026-09-17", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies how automatic defaults and reflective social-influence cues affect GUI-agent shopping decisions under different reasoning configurations. A randomized simulated-shopping experiment finds that reasoning changes the type of susceptibility rather than uniformly removing interface influence.", "topics": ["safety-security", "evaluation", "ui-design", "robustness"], "updated_at": "2026-09-17", "url": "https://arxiv.org/abs/2609.19843"}
{"aliases": [], "authors": ["Zeyu Kang", "Zhenyun Yin", "Yang Zhang", "Shan He", "Shanzhe Lei", "Yanjiu Zhong", "Xinquan Chen", "Xuhong Wang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "a88d287d5433009962a905dbdbf03359e1183e66e4a6015441ec28405f64a6ee", "submitted_at": "2026-08-27T05:43:50Z", "updated_at": "2026-09-22T06:39:18Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2609.22178v2"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.22178", "identifiers": {"arxiv": "2609.22178"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.22178"}], "name": "Beyond Task Completion: Training Capable and Safe Computer-Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-08-27", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces SCOPE, which trains computer-use agents on ordinary task demonstrations, safe continuations around hazards, and explicit refusals. Its SCOPE-Gen pipeline pairs verifiable tasks with environmental-risk variants to support joint capability and safety training.", "topics": ["safety-security", "reinforcement-learning", "data-generation"], "updated_at": "2026-09-22", "url": "https://arxiv.org/abs/2609.22178"}
{"aliases": [], "authors": ["Yuxuan Li", "Will Epperson", "Wesley Deng", "Zezhou Huang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "8d241c1ea9731bc815eef5df9c409d410608adb052616406d955d7d33cecd9b9", "submitted_at": "2026-09-23T03:00:00Z", "updated_at": "2026-09-27T20:49:01Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2609.27273v2"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.27273", "identifiers": {"arxiv": "2609.27273"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.27273"}], "name": "CAVEAT: Towards Robust Computer-Use Agents in Incentive-Misaligned Environments", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-09-23", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces CAVEAT, a controlled marketplace benchmark for testing whether delegated agents preserve user objectives when interfaces favor other outcomes. It diagnoses decision failures and studies harness interventions and targeted training to improve resistance to marketplace steering.", "topics": ["evaluation", "robustness", "safety-security"], "updated_at": "2026-09-27", "url": "https://arxiv.org/abs/2609.27273"}
{"aliases": [], "authors": ["Zeyu Zhang", "Dingyi Rong", "Zijian Chen", "Zicheng Zhang", "Xiongkuo Min", "Guangtao Zhai"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "3a89489bb1ff3001a1d806c3eefd8a6aa19503d1dfbe0265b916766119f96ec2", "submitted_at": "2026-09-23T07:27:23Z", "updated_at": "2026-09-23T07:27:23Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.27461v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.27461", "identifiers": {"arxiv": "2609.27461"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.27461"}], "name": "Invisible in Space, Visible in Time: Motion Vision CAPTCHA against GUI Agents", "notes": "The annotation was reviewed against the complete retained official abstract. Experimental findings and artifact availability have not been independently reproduced.", "published_at": "2026-09-23", "relations": [], "review_status": "curated", "scope": "core", "summary": "MVCAP-Bench tests live browser CAPTCHAs whose foreground meaning emerges through motion against dynamic backgrounds. A foreground-only control separates temporal camouflage difficulty from answer format and browser interaction.", "topics": ["evaluation", "screen-understanding", "video", "safety-security"], "updated_at": "2026-09-23", "url": "https://arxiv.org/abs/2609.27461"}
{"aliases": [], "authors": ["Asif Shahriar", "Md Nafiu Rahman", "Sadif Ahmed", "Farig Sadeque", "Md Rizwan Parvez"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "802342ce2c486c30c84c33c897becb0348831034d4d11f8d3d3861ec0380b65c", "submitted_at": "2026-09-26T20:09:16Z", "updated_at": "2026-09-26T20:09:16Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.32915v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:33:17Z", "role": "primary", "sha256": "5d7c9ff24833b254d33a41a7c4247e830e4d9c371881a1e16151bad5c71cc8b3", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2609.10297%2C2609.10451%2C2609.12394%2C2609.12404%2C2609.13287%2C2609.13491%2C2609.13548%2C2609.13873%2C2609.14239%2C2609.15134%2C2609.15307%2C2609.15364%2C2609.15457%2C2609.16251%2C2609.16635%2C2609.16732%2C2609.16900%2C2609.17653%2C2609.17885%2C2609.18411%2C2609.18690%2C2609.19125%2C2609.19523%2C2609.19843%2C2609.22083%2C2609.22724%2C2609.24277%2C2609.24890%2C2609.25769%2C2609.27273%2C2609.27307%2C2609.27353%2C2609.27461%2C2609.29892%2C2609.30186%2C2609.30604%2C2609.30887%2C2609.31957%2C2609.32036%2C2609.32130%2C2609.32166%2C2609.32351%2C2609.32385%2C2609.32600%2C2609.32679%2C2609.32750%2C2609.32915%2C2609.33304%2C2609.33646%2C2609.34113&max_results=50"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:33:17Z", "role": "primary", "sha256": "5d7c9ff24833b254d33a41a7c4247e830e4d9c371881a1e16151bad5c71cc8b3", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2609.10297%2C2609.10451%2C2609.12394%2C2609.12404%2C2609.13287%2C2609.13491%2C2609.13548%2C2609.13873%2C2609.14239%2C2609.15134%2C2609.15307%2C2609.15364%2C2609.15457%2C2609.16251%2C2609.16635%2C2609.16732%2C2609.16900%2C2609.17653%2C2609.17885%2C2609.18411%2C2609.18690%2C2609.19125%2C2609.19523%2C2609.19843%2C2609.22083%2C2609.22724%2C2609.24277%2C2609.24890%2C2609.25769%2C2609.27273%2C2609.27307%2C2609.27353%2C2609.27461%2C2609.29892%2C2609.30186%2C2609.30604%2C2609.30887%2C2609.31957%2C2609.32036%2C2609.32130%2C2609.32166%2C2609.32351%2C2609.32385%2C2609.32600%2C2609.32679%2C2609.32750%2C2609.32915%2C2609.33304%2C2609.33646%2C2609.34113&max_results=50"}], "id": "arxiv:2609.32915", "identifiers": {"arxiv": "2609.32915"}, "kind": "paper", "last_verified_at": "2026-10-16T06:33:17Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.32915"}], "name": "AgentTell: Behavioural Side-Channel Leakage in Browser-Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-09-26", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces AgentTell, a benchmark for privacy leakage through browser actions after an agent learns a private fact on another website. It examines whether agents choose revealing options despite nondisclosure instructions and whether their final reports accurately describe that behavior.", "topics": ["safety-security", "evaluation", "memory"], "updated_at": "2026-09-26", "url": "https://arxiv.org/abs/2609.32915"}
{"aliases": [], "authors": ["Ben Hagag", "William L. Anderson", "Srija Chakraborty", "Christian Schroeder de Witt"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "3c380079220050fbe354509238a5a79c7df2917a245490e4ec0cff512f7565ab", "submitted_at": "2026-09-27T02:20:08Z", "updated_at": "2026-09-27T02:20:08Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.33102v1"}}, "environments": ["web", "desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-20T12:58:20.217517Z", "role": "primary", "sha256": "f275c47a8b6d349a70e39a9e55eea5a09d99d5f35cc896100f3bcfd5b0923e0b", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28abs%3A%22computer-use+agents%22+OR+abs%3A%22GUI+agents%22+OR+abs%3A%22browser+agents%22+OR+abs%3A%22mobile+GUI%22%29+AND+submittedDate%3A%5B202609200000+TO+202610201255%5D&start=0&max_results=200&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T13:12:28.833839+00:00", "role": "primary", "sha256": "3ea667f4187ba601011e5ff00c4d63676b6d9903d12b80a0d7d5dd9920f00c65", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://arxiv.org/html/2609.33102v1"}], "id": "arxiv:2609.33102", "identifiers": {"arxiv": "2609.33102"}, "kind": "paper", "last_verified_at": "2026-10-20T13:12:28.833839+00:00", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.33102"}], "name": "ORBIT: A Framework for Multi-Agent Safety and Security Evaluations", "notes": "BrowserART uses an accessibility-tree action interface, while the desktop adapter uses mouse, keyboard and screenshots. The adapted tasks are not claimed to preserve the constructs of their source benchmarks. Experiments cover a representative configuration slice with fixed attack templates; the OS-Harm results are imported from prior work using a different model and stack. Browser defense tables also differ in baseline versus defended judges and in judged benign denominators. Human judge validation used one grader. These results do not establish a universally effective defense. The linked repository and its code/data licenses were not independently inspected, and no experiment was run.", "published_at": "2026-09-27", "relations": [], "review_status": "curated", "scope": "core", "summary": "Inspect-based framework that varies multi-agent roles, communication, attacks and defenses, with BrowserGym and virtual-desktop GUI adapters alongside coding and other tool scenarios.", "topics": ["evaluation", "safety-security", "web-navigation"], "updated_at": "2026-09-27", "url": "https://arxiv.org/abs/2609.33102"}
{"aliases": [], "authors": ["Ruozhao Yang", "Mingfei Cheng", "Xiaofei Xie"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "a9bc3b57957e6c4f1af399379c4275366563b94a8b8fe01ec83e54715cfc8d11", "submitted_at": "2026-09-28T11:53:24Z", "updated_at": "2026-09-28T11:53:24Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.34974v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.34974", "identifiers": {"arxiv": "2609.34974"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.34974"}], "name": "Before Acting, Change the State: Prospective State Intervention for Web Agents under Deceptive Interfaces", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-09-28", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces Veer, a runtime defense that changes task-relevant web state when an otherwise valid proposed action would produce an unauthorized consequence. It plans a prospective intervention and grounds and verifies its execution before the base agent continues.", "topics": ["safety-security", "planning", "robustness"], "updated_at": "2026-09-28", "url": "https://arxiv.org/abs/2609.34974"}
{"aliases": [], "authors": ["Jianxing Chen", "Xiao Yu", "Shipra Agrawal", "Zhou Yu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e92c63d4cf58375aa7068570f8a98c145e24f8d393c5807c0aff3ed67f18b9f0", "submitted_at": "2026-09-28T20:01:35Z", "updated_at": "2026-09-28T20:01:35Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.36201v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.36201", "identifiers": {"arxiv": "2609.36201"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.36201"}], "name": "SCOUT: Synergizing Reasoning and Tool-Use for Computer-Use Safety", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-09-28", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces SCOUT, a safety verifier that first derives task-specific completion and safety rubrics from a computer-use trajectory, then actively probes the resulting environment for evidence. It evaluates whether these two stages detect harmful consequences that screenshots alone can miss.", "topics": ["safety-security", "evaluation", "tool-integration"], "updated_at": "2026-09-28", "url": "https://arxiv.org/abs/2609.36201"}
{"aliases": [], "authors": ["Yongjiang Wu", "Junyuan Zhang", "Ada Chen", "Kuiyi Gao", "Wenxuan Wang"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "f14844abefab6b4eb464a1c7675f9de7f460849872ebb776230724ed5828feed", "submitted_at": "2026-08-07T07:51:13Z", "updated_at": "2026-08-07T07:51:13Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2609.38184v1"}}, "environments": ["web", "desktop", "cross-platform"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:32Z", "role": "primary", "sha256": "13d3fb5811d203f2e6b80eb45d30ebc0b3cfefd88e2034ffa3b1b80687bef285", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=100&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2609.38184", "identifiers": {"arxiv": "2609.38184"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:32Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2609.38184"}], "name": "RealGUINoise: An Interactive Cross-Platform Benchmark for GUI Agent Robustness under Real-World Interface Noise", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-08-07", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates interactive GUI agents under varied realistic interface noise, comparing task completion, safety, and behavior with clean-environment runs.", "topics": ["robustness", "evaluation", "safety-security"], "updated_at": "2026-08-07", "url": "https://arxiv.org/abs/2609.38184"}
{"aliases": [], "authors": ["Giulio Zingrillo", "Hanna Foerster", "Ilia Shumailov", "Yiren Zhao", "Robert Mullins"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "7469484bb1186dd633b26e41dddc890008e893c303f93d2352f93c9d47fd78c2", "submitted_at": "2026-10-02T10:08:39Z", "updated_at": "2026-10-02T10:08:39Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.03089v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2610.03089", "identifiers": {"arxiv": "2610.03089"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.03089"}], "name": "Securing Computer-Use Agents Against Branch Steering Attacks", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-02", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces STEER-Bench to study branch-steering attacks that manipulate untrusted web data toward hazardous paths already permitted by an agent’s plan. Its COBRA architecture combines trusted branching plans with advance constraints on the parameters and destinations each branch can use.", "topics": ["safety-security", "planning", "evaluation"], "updated_at": "2026-10-02", "url": "https://arxiv.org/abs/2610.03089"}
{"aliases": [], "authors": ["Maria Sanchez del Rio", "Mikhail Terekhov"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "f9dd28bdfb8879cbdb68f539c286365a06f4e9d04eba2420bb46f3575fdbfa05", "submitted_at": "2026-10-02T00:31:34Z", "updated_at": "2026-10-12T23:17:28Z", "version": 2, "versioned_url": "https://arxiv.org/abs/2610.03818v2"}}, "environments": ["desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2610.03818", "identifiers": {"arxiv": "2610.03818"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.03818"}], "name": "Control OSWorld: An AI Control Environment for GUI Computer Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-02", "relations": [], "review_status": "curated", "scope": "core", "summary": "Pairs OSWorld tasks with harmful side objectives to evaluate monitors that detect malicious behavior in computer-use agent trajectories.", "topics": ["safety-security", "evaluation"], "updated_at": "2026-10-12", "url": "https://arxiv.org/abs/2610.03818"}
{"aliases": [], "authors": ["Li Hu", "Kanghua Mo", "Yingbin Jin", "Qingqing Ye", "Haibo Hu"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "5afea9414462386ee921b9ead9381c0204963c710c9f616a96f4bc47079f059e", "submitted_at": "2026-09-29T03:03:00Z", "updated_at": "2026-09-29T03:03:00Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.06898v1"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-18T10:00:31.520084+00:00", "role": "primary", "sha256": "f053c4c65f6443acbda26e25cf36f7b6e8ea45f2916535604229f7c4e62ed679", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22+OR+ti%3A%22mobile+agent%22+OR+ti%3A%22UI+agent%22+OR+ti%3A%22desktop+agent%22+OR+ti%3A%22screen+grounding%22%29+AND+submittedDate%3A%5B200001010000+TO+202610180000%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:00:31.520084+00:00", "role": "primary", "sha256": "f053c4c65f6443acbda26e25cf36f7b6e8ea45f2916535604229f7c4e62ed679", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22+OR+ti%3A%22mobile+agent%22+OR+ti%3A%22UI+agent%22+OR+ti%3A%22desktop+agent%22+OR+ti%3A%22screen+grounding%22%29+AND+submittedDate%3A%5B200001010000+TO+202610180000%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:10:35.098754+00:00", "role": "primary", "sha256": "73f83c27e58e46086dbd5f32f108df2c5faa9ab2bd968d48062f39d94b2cc14b", "supports": ["summary", "scope", "environments", "notes"], "url": "https://arxiv.org/pdf/2610.06898v1"}], "id": "arxiv:2610.06898", "identifiers": {"arxiv": "2610.06898"}, "kind": "paper", "last_verified_at": "2026-10-18T10:00:31.520084+00:00", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.06898"}], "name": "DIBench: Benchmarking Decision Integrity of GUI-based Mobile Agents Under Deceptive Injections", "notes": "The commercial-app protocol uses fixed screenshots and context for selection decisions, rather than live end-to-end transactions. Android is supported by the paper's emulator setup. Completion and integrity are distinct measures; no reported scores were reproduced. Basis: complete retained abstract and source-reviewed sections of the exact-version paper (3.1 Benchmark components; 3.2 Instance generation; 3.3 Paired protocol/metrics; 5 Limitations/Ethics; Appendix B (Android emulator setup)); not an entire-paper reading or reproduction claim.", "published_at": "2026-09-29", "relations": [], "review_status": "curated", "scope": "core", "summary": "Evaluates whether deceptive UI content steers mobile agents away from instruction-consistent choices, using paired benign and injected decision states and decision-integrity metrics.", "topics": ["evaluation", "safety-security", "mobile-navigation"], "updated_at": "2026-09-29", "url": "https://arxiv.org/abs/2610.06898"}
{"aliases": [], "authors": ["Sarim Hashmi", "Mukul Ranjan", "Kshitij Mishra", "Mikhail Kuznetsov", "Praneeth Vepakomma", "Nils Lukas"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "7113025e0dfd784d6994108c24681000625b13598e33ad0e4324e90fd2832c19", "submitted_at": "2026-10-06T17:56:43Z", "updated_at": "2026-10-06T17:56:43Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.08773v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2610.08773", "identifiers": {"arxiv": "2610.08773"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.08773"}], "name": "AdvSim2Real : Training Web Agents Against Adaptive Prompt Injection in a Web World Model", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-06", "relations": [], "review_status": "curated", "scope": "core", "summary": "Presents AdvSim2Real, which jointly adapts web-task curricula, prompt-injection adversaries, and agent policies inside a frozen web world model. It evaluates robustness against an unseen adversary and transfer from simulated interaction to a real browser.", "topics": ["safety-security", "world-models", "reinforcement-learning"], "updated_at": "2026-10-06", "url": "https://arxiv.org/abs/2610.08773"}
{"aliases": [], "authors": ["Wanjing Han", "Levi Taiji Li", "Mu Zhang", "Yue Jiang", "Guanhong Tao"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "3b6ef7e785ad509ebd13fdb3da97fc583ee4c28dd42b4eab842c6cd88d7ef51f", "submitted_at": "2026-10-07T00:05:53Z", "updated_at": "2026-10-07T00:05:53Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.09240v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2610.09240", "identifiers": {"arxiv": "2610.09240"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.09240"}], "name": "Adversarial Images Hijack Web Agents from Visual Grounding to Browser Execution", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-07", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces WebMirage, a visual red-teaming framework that models webpage element competition and action post-processing to test whether manipulated grounding decisions lead to executed browser actions.", "topics": ["safety-security", "grounding", "evaluation"], "updated_at": "2026-10-07", "url": "https://arxiv.org/abs/2610.09240"}
{"aliases": [], "authors": ["Sarthak Choudhary", "Mihai Christodorescu", "Ashish Hooda", "Somesh Jha", "Tongxin Li", "Damien Octeau"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "f38c9e37ab7fc8356a0c8c72370de673d1c2b6294e15dfdc72c280269f43439c", "submitted_at": "2026-10-07T05:24:40Z", "updated_at": "2026-10-07T05:24:40Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.09469v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2610.09469", "identifiers": {"arxiv": "2610.09469"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.09469"}], "name": "Secure-CUA: Controlling Untrusted Influence in Computer-Use Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-07", "relations": [], "review_status": "curated", "scope": "core", "summary": "Proposes Secure-CUA, which commits to per-action transactions before reading untrusted page content and separates content queries from GUI target selection. The paper formalizes its protection assumptions and evaluates benign task utility on WebArena.", "topics": ["safety-security", "web-navigation"], "updated_at": "2026-10-07", "url": "https://arxiv.org/abs/2610.09469"}
{"aliases": [], "authors": ["Jiaming Zhang", "Xuan Wang", "Fuyao Zhang", "Yang Cao", "Lingjuan Lyu", "Wei Yang Bryan Lim"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "53120f960c7b98e87210c93ac661a2e397b858cc9f211786b0f8233fa61c6dd3", "submitted_at": "2026-10-08T13:33:23Z", "updated_at": "2026-10-08T13:33:23Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.11942v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2610.11942", "identifiers": {"arxiv": "2610.11942"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.11942"}], "name": "Right Screen, Wrong Transition: World Models as Verifiers for GUI Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-08", "relations": [], "review_status": "curated", "scope": "core", "summary": "Presents LGWM, a decoder-free GUI world model that predicts the next screen representation from an action and compares it with the observed state to verify transitions. Its paired diagnostic separates legitimate and hijacked transitions that lead to the same credential screen.", "topics": ["world-models", "safety-security", "evaluation"], "updated_at": "2026-10-08", "url": "https://arxiv.org/abs/2610.11942"}
{"aliases": [], "authors": ["Md Imran Hossen"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "63e19265f1df2c336e28c4a13e11331021bd60866269c0b921d92a3c0fec080d", "submitted_at": "2026-10-10T19:30:31Z", "updated_at": "2026-10-10T19:30:31Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.15138v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2610.15138", "identifiers": {"arxiv": "2610.15138"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.15138"}], "name": "The Agent Clicks Approve: Injected Page Text Makes Browser Agents Authorize Access the Task Never Named", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-10", "relations": [], "review_status": "curated", "scope": "core", "summary": "Studies prompt-injected account-access approvals in browser agents and proposes a service-side authority envelope that checks authorization requests against the user task before allowing them to complete.", "topics": ["safety-security", "evaluation"], "updated_at": "2026-10-10", "url": "https://arxiv.org/abs/2610.15138"}
{"aliases": [], "authors": ["Hyobin Ahn", "Dongjae Lee"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "ff5470e792d14ced2cdff648d97e2606eea4b3c822046b7c10272fd2a5f1695f", "submitted_at": "2026-10-11T10:09:44Z", "updated_at": "2026-10-11T10:09:44Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.15502v1"}}, "environments": ["android", "web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-20T12:56:03.829575Z", "role": "primary", "sha256": "a8636ac679d2110181d4417d4aef040ddb24464597a7fa6fe6d2c35a2401a3b1", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22graphical+user+interface%22+OR+ti%3A%22screen+grounding%22+OR+ti%3A%22UI+grounding%22+OR+ti%3A%22desktop+agents%22+OR+ti%3A%22Android+agent%22+OR+ti%3A%22phone+agent%22+OR+ti%3A%22mobile+GUI%22+OR+ti%3A%22web+navigation%22+OR+ti%3A%22GUI+automation%22+OR+ti%3A%22browser+automation%22%29+AND+submittedDate%3A%5B202608210000+TO+202610201255%5D&start=0&max_results=200&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T13:12:58.069698Z", "role": "primary", "sha256": "a38b3ddc9d5d6744eefc937f86e174917f9b5de149323632d5d1b20283d79c8e", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://arxiv.org/html/2610.15502v1"}], "id": "arxiv:2610.15502", "identifiers": {"arxiv": "2610.15502"}, "kind": "paper", "last_verified_at": "2026-10-20T13:12:58.069698Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.15502"}], "name": "From Observations to Authorized Actions: Measuring Source Information in Android Agents", "notes": "The two prototypes use different workloads and trust assumptions on patched Android/WebView components. Approval is simulated; authenticated browser delivery and atomic capture remain unimplemented. Coarse source history can withhold benign work and persist across tasks. Direct-call checks and limited live-agent trials are separate, and writes are in-memory. No production vulnerability, general defense rate or released artifact is established.", "published_at": "2026-10-11", "relations": [], "review_status": "curated", "scope": "core", "summary": "Measures source-attribution limits in Android accessibility and separately tests application-history and selected-WebView-object checks for authorizing AppFunctions actions.", "topics": ["safety-security", "tool-integration", "observability"], "updated_at": "2026-10-11", "url": "https://arxiv.org/abs/2610.15502"}
{"aliases": [], "authors": ["Zian Yuan", "Wei Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "c5a2743a18bae1deeed13fbd1dd7379c3f21c67488627057ccb466ffe51d09dc", "submitted_at": "2026-10-12T13:34:52Z", "updated_at": "2026-10-12T13:34:52Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.16778v1"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2610.16778", "identifiers": {"arxiv": "2610.16778"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.16778"}], "name": "PrivacyGuard: Source-Linked Sanitization and Locally Resolvable PII References for Cloud GUI Agents", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-12", "relations": [], "review_status": "curated", "scope": "core", "summary": "Introduces PrivacyGuard, which sanitizes source-linked personal information and maintains locally resolvable references across mobile-agent turns to reduce exposure to remote planners.", "topics": ["safety-security", "mobile-navigation", "evaluation"], "updated_at": "2026-10-12", "url": "https://arxiv.org/abs/2610.16778"}
{"aliases": [], "authors": ["Yuetian Chen", "Kaiyuan Zhang", "Yuntao Du", "Thomas Li", "Mark Tenenholtz", "Kyle Polley", "Jerry Ma", "Denis Yarats", "Ninghui Li"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "247cd4c9634dea78780a2230138c728adf53871e8b65614697f76ba53dea858c", "submitted_at": "2026-10-12T22:17:51Z", "updated_at": "2026-10-12T22:17:51Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.17457v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:33:24Z", "role": "primary", "sha256": "9641755401e0e8ec9613f7c4509029d968ab4f10751ba4d43cab57e7d27c1da1", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2610.11655%2C2610.11707%2C2610.11942%2C2610.12611%2C2610.12823%2C2610.12866%2C2610.12984%2C2610.13268%2C2610.13416%2C2610.13703%2C2610.14157%2C2610.14228%2C2610.15138%2C2610.15765%2C2610.16234%2C2610.16319%2C2610.16720%2C2610.16778%2C2610.17457%2C2610.17710%2C2610.17792%2C2610.17895%2C2610.17928%2C2610.18508%2C2610.18784%2C2610.18868%2C2610.19100%2C2610.19128%2C2610.20584&max_results=29"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:33:24Z", "role": "primary", "sha256": "9641755401e0e8ec9613f7c4509029d968ab4f10751ba4d43cab57e7d27c1da1", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?id_list=2610.11655%2C2610.11707%2C2610.11942%2C2610.12611%2C2610.12823%2C2610.12866%2C2610.12984%2C2610.13268%2C2610.13416%2C2610.13703%2C2610.14157%2C2610.14228%2C2610.15138%2C2610.15765%2C2610.16234%2C2610.16319%2C2610.16720%2C2610.16778%2C2610.17457%2C2610.17710%2C2610.17792%2C2610.17895%2C2610.17928%2C2610.18508%2C2610.18784%2C2610.18868%2C2610.19100%2C2610.19128%2C2610.20584&max_results=29"}], "id": "arxiv:2610.17457", "identifiers": {"arxiv": "2610.17457"}, "kind": "paper", "last_verified_at": "2026-10-16T06:33:24Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.17457"}], "name": "BrowseSafe++: Benchmarking Prompt Injection Detection Across Web Contexts, Languages, and Browser Views", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-12", "relations": [], "review_status": "curated", "scope": "core", "summary": "Extends BrowseSafe with paired attacked and clean webpages in several representations, and studies text and screenshot detectors plus feedback-adapted prompt injections.", "topics": ["safety-security", "evaluation", "web-navigation"], "updated_at": "2026-10-12", "url": "https://arxiv.org/abs/2610.17457"}
{"aliases": [], "authors": ["Dominik Seip", "Bartol Markovinović", "Matthias Hein"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "5565d38dd40b02b4158d7edcc2c5c27e7e285dbedefe1c7101a31c4f8d17b659", "submitted_at": "2026-10-13T15:03:33Z", "updated_at": "2026-10-13T15:03:33Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.18508v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2610.18508", "identifiers": {"arxiv": "2610.18508"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.18508"}], "name": "Towards Safe Autonomy for Computer Use", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-13", "relations": [], "review_status": "curated", "scope": "core", "summary": "Presents SA-CUA, separating trusted and untrusted visual content using accessibility-derived masking rules and a controlled channel for verifying information before endorsement.", "topics": ["safety-security", "screen-understanding"], "updated_at": "2026-10-13", "url": "https://arxiv.org/abs/2610.18508"}
{"aliases": [], "authors": ["Yuke Wang", "Benlei Cui", "Shen Pang", "Xuemei Dong", "Longtao Huang", "Hui Xue", "Yuwen Zhai", "Junjie Li", "Jingqun Tang", "Haiwen Hong"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "da334d943d98b4bf9915c6ca28aab734d736f376e67cbab950aa1af65f540ebf", "submitted_at": "2026-10-14T01:30:03Z", "updated_at": "2026-10-14T01:30:03Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.19384v1"}, "license_review": {"document": {"evidence_urls": ["https://arxiv.org/html/2610.19384v1"], "reason": "Paper-specific document declaration only; no code, data or weights permissions are inferred."}}}, "environments": ["web", "desktop"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-20T05:52:01Z", "role": "primary", "sha256": "63e37cd26bfe69455a3fa835f3d6ba4cfcc5bb7dc91f90505920d37b1e08f0b6", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2610.19384%2C2610.22672%2C2610.22697%2C2610.23303%2C2610.23307%2C2610.24729&max_results=6"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T19:02:16.329923+00:00", "role": "primary", "sha256": "ca995ec330f62a43d3d4fbbf28dfe83c5fa8b05f5c1e446929493409760268d1", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes", "licenses.document"], "url": "https://arxiv.org/html/2610.19384v1"}], "id": "arxiv:2610.19384", "identifiers": {"arxiv": "2610.19384"}, "kind": "paper", "last_verified_at": "2026-10-20T19:02:16.329923+00:00", "licenses": {"code": null, "data": null, "document": {"basis": "The retained exact-v1 arXiv HTML infobox explicitly declares License: CC BY 4.0 for this document.", "declared": "CC-BY-4.0", "retrieved_at": "2026-10-20T19:02:16.329923+00:00", "url": "https://arxiv.org/html/2610.19384v1"}, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.19384"}], "name": "MMA-SafetyBench: A Benchmark for Multimodal Agent Safety Evaluation", "notes": "ASR measures predicted coordinates inside an annotated target or compromised generated content; actual clicks, environment changes and downstream harm are not verified. SCR additionally counts adopted adversarial intent or task/schema disruption. Clean-task and attacked-trajectory denominators differ, including 42 clean desktop tasks versus 209 adversarial desktop trajectories. Payloads and injection points are fixed, and neutral prompts change both compliance cues and instruction-priority guidance. GPT-5 judging has a 300-trajectory human-label check, with borderline disagreements. Android-style mobile dialogs do not establish an Android runtime. Code/data links were identified but not inspected; no independent reproduction.", "published_at": "2026-10-14", "relations": [], "review_status": "curated", "scope": "core", "summary": "Builds trajectory-guided visual-injection tests for web, desktop and mobile observations, separating target-matching action predictions from broader semantic compromise.", "topics": ["evaluation", "safety-security", "grounding"], "updated_at": "2026-10-14", "url": "https://arxiv.org/abs/2610.19384"}
{"aliases": [], "authors": ["Wenkui Yang", "Yiming Hao", "Kai Zhou", "Chengyu Pu", "Yuzhe Zhang", "Jie Cao", "Jin Huang", "Shuai Ma", "Feng Tian"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "602f4eb7aacc05748b83b9c32b28439098005c77eafb7624804f820995521a79", "submitted_at": "2026-10-14T12:35:59Z", "updated_at": "2026-10-14T12:35:59Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.20100v1"}}, "environments": ["unspecified"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:29Z", "role": "primary", "sha256": "74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=0&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "arxiv:2610.20100", "identifiers": {"arxiv": "2610.20100"}, "kind": "paper", "last_verified_at": "2026-10-16T06:36:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.20100"}], "name": "Task-State Governance: A Virtual Machine for Verifiable and Recoverable GUI Agent Execution", "notes": "The annotation and scope were reviewed against the primary paper abstract. Experimental results and released artifacts have not been independently reproduced.", "published_at": "2026-10-14", "relations": [], "review_status": "curated", "scope": "core", "summary": "Presents TaskVM, an interface for specifying desired and preserved application state, checking agent actions, and exposing whether changes can be reversed or compensated.", "topics": ["human-in-the-loop", "safety-security", "workflow"], "updated_at": "2026-10-14", "url": "https://arxiv.org/abs/2610.20100"}
{"aliases": [], "authors": ["Umar Iqbal"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "077dd623c980813e83197a38b6545161a3d2fc842d99fe291c53372b9770acd2", "submitted_at": "2026-10-15T22:02:45Z", "updated_at": "2026-10-15T22:02:45Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.22697v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-20T05:52:01Z", "role": "primary", "sha256": "63e37cd26bfe69455a3fa835f3d6ba4cfcc5bb7dc91f90505920d37b1e08f0b6", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2610.19384%2C2610.22672%2C2610.22697%2C2610.23303%2C2610.23307%2C2610.24729&max_results=6"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T13:38:46.444001Z", "role": "primary", "sha256": "cb6a8a9c0cc47be21b36b3d1831b26c6189d8f2e4af0d7e26d643c9c2a65452f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes"], "url": "https://arxiv.org/html/2610.22697v1"}], "id": "arxiv:2610.22697", "identifiers": {"arxiv": "2610.22697"}, "kind": "paper", "last_verified_at": "2026-10-20T13:38:46.444001Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.22697"}], "name": "What It Is, Not What It Says: Securing Agentic Browsing Interactions with Web Semantics", "notes": "Evaluation instructs the model to simulate compromise and measures planner outputs and permission appearance. Utility uses plan matching and LLM-judged element suitability; no steps are executed in that assessment. Assumes an honest first party, trusted browser/harness and untampered instrumentation; multi-page and availability attacks are outside scope. Parallel latency reduction is an unimplemented estimate. No independent reproduction.", "published_at": "2026-10-15", "relations": [], "review_status": "curated", "scope": "core", "summary": "Intact records webpage element provenance and capabilities, isolates planning per element, and uses a browser harness to constrain actions and request consent for third-party interactions.", "topics": ["safety-security", "web-navigation", "planning", "human-in-the-loop"], "updated_at": "2026-10-15", "url": "https://arxiv.org/abs/2610.22697"}
{"aliases": [], "authors": ["Songping Wang", "Borui Li", "Qinglong Liu", "Yueming Lyu", "Yi Huang", "Fengting Li", "Chao Feng", "Caifeng Shan"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "ea753c1cb0666950d89f2781511f26872e3d5b1c2fd2434896248214160fa22c", "submitted_at": "2026-10-16T11:06:37Z", "updated_at": "2026-10-16T11:06:37Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.23303v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-20T05:52:01Z", "role": "primary", "sha256": "63e37cd26bfe69455a3fa835f3d6ba4cfcc5bb7dc91f90505920d37b1e08f0b6", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2610.19384%2C2610.22672%2C2610.22697%2C2610.23303%2C2610.23307%2C2610.24729&max_results=6"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T08:27:26.236797+00:00", "role": "primary", "sha256": "80bed80786f939973d0d13b4f892300fee02a6110e326247131fc83d257796de", "supports": ["name", "kind", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agent%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22+OR+ti%3A%22browser+agents%22+OR+ti%3A%22mobile+agent%22+OR+ti%3A%22mobile+agents%22+OR+ti%3A%22desktop+agent%22%29+AND+submittedDate%3A%5B202610130000+TO+202610192359%5D&start=0&max_results=200&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T10:42:09.967883+00:00", "role": "primary", "sha256": "ec9d647818be346ddbf7807be87b4789cd53e50c5d3d7a7904ad4f49bc71c38e", "supports": ["summary", "environments", "topics", "scope", "notes"], "url": "https://arxiv.org/html/2610.23303v1"}], "id": "arxiv:2610.23303", "identifiers": {"arxiv": "2610.23303"}, "kind": "paper", "last_verified_at": "2026-10-20T08:27:26.236797+00:00", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.23303"}], "name": "MobileJailBench: Red Queen Jailbreak Benchmarking for Mobile Agents", "notes": "Reviewed the complete retained abstract and selected exact-v1 threat-model, construction and evaluation sections. Execution uses a browser-based Android-like simulator, not real Android isolation or live services. The attacker controls instructions and designated carriers, not privileged simulator state/APIs or system/model/harness internals. Rates depend on release, budget and adjudication coverage. SAF labels track observable expressions and actions, not latent reasoning; unresolved episodes are excluded from rate denominators. No independent reproduction.", "published_at": "2026-10-16", "relations": [], "review_status": "curated", "scope": "core", "summary": "MobileJailBench evaluates mobile-agent jailbreaks in a browser-based Android-like simulator, combining feedback-driven attack evolution, separately admitted task expansions and evidence-grounded outcome adjudication.", "topics": ["safety-security", "mobile-navigation", "evaluation", "observability", "cross-app"], "updated_at": "2026-10-16", "url": "https://arxiv.org/abs/2610.23303"}
{"aliases": [], "authors": ["Songping Wang", "Peiyang Li", "Yi Huang", "Fengting Li", "Xuhao Ran", "Yueming Lyu", "Chao Feng", "Caifeng Shan"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "3fc22fbeda19ac2947b00af73801f0674c550e0d7c78919f811cae754b06128c", "submitted_at": "2026-10-16T11:10:30Z", "updated_at": "2026-10-16T11:10:30Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.23307v1"}}, "environments": ["android"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-20T05:52:01Z", "role": "primary", "sha256": "63e37cd26bfe69455a3fa835f3d6ba4cfcc5bb7dc91f90505920d37b1e08f0b6", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?id_list=2610.19384%2C2610.22672%2C2610.22697%2C2610.23303%2C2610.23307%2C2610.24729&max_results=6"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T08:27:26.236797+00:00", "role": "primary", "sha256": "80bed80786f939973d0d13b4f892300fee02a6110e326247131fc83d257796de", "supports": ["name", "kind", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agent%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22+OR+ti%3A%22browser+agents%22+OR+ti%3A%22mobile+agent%22+OR+ti%3A%22mobile+agents%22+OR+ti%3A%22desktop+agent%22%29+AND+submittedDate%3A%5B202610130000+TO+202610192359%5D&start=0&max_results=200&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T09:17:36.223826+00:00", "role": "primary", "sha256": "3803984a185b6258276b87b87ac1991583878fb41dc22618ff743f6f69c7bc7e", "supports": ["summary", "environments", "topics", "scope", "notes"], "url": "https://arxiv.org/html/2610.23307v1"}], "id": "arxiv:2610.23307", "identifiers": {"arxiv": "2610.23307"}, "kind": "paper", "last_verified_at": "2026-10-20T08:27:26.236797+00:00", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.23307"}], "name": "MobileInject: Evaluating Reasoning-to-Action Security of MLLM-Based Mobile Agents under Cross-Modal Consistency Injection", "notes": "Reviewed the complete abstract and selected v1 construction, threat-model and evaluation sections. The attacker modifies observation surfaces, not system instructions, model parameters or the OS. State oracles are distinct from emitted-rationale diagnoses, whose labels include outcome-dependent constraints. Those diagnostics are not independent evidence of hidden-reasoning causality. No reproduction or release verification.", "published_at": "2026-10-16", "relations": [], "review_status": "curated", "scope": "core", "summary": "MobileInject evaluates visual/textual observation injection in stateful Android workflows using resettable emulators, benign/adversarial state oracles and trace-based diagnosis of emitted rationales and actions.", "topics": ["safety-security", "mobile-navigation", "evaluation", "observability", "cross-app"], "updated_at": "2026-10-16", "url": "https://arxiv.org/abs/2610.23307"}
{"aliases": [], "authors": ["Egor Zverev", "Juhee Kim", "Leon Trampert", "Boris Köpf", "Aashish Kolluri", "Shruti Tople", "Santiago Zanella-Béguelin"], "details": {"arxiv_snapshot": {"abstract_hash_method": "sha256-utf8-whitespace-collapsed-v1", "abstract_sha256": "e675a3082f571da547fb5b6cb89e4f68a034bfa57236886e1218cb2bb2cdf1c3", "submitted_at": "2026-10-16T22:12:45Z", "updated_at": "2026-10-16T22:12:45Z", "version": 1, "versioned_url": "https://arxiv.org/abs/2610.24277v1"}}, "environments": ["web"], "evidence": [{"method": "arxiv_api", "retrieved_at": "2026-10-20T05:52:00Z", "role": "primary", "sha256": "af310824eea97aac8209023ff53bad7bea77778816586b10bb94110148347af9", "supports": ["name", "authors", "published_at", "updated_at", "identifiers.arxiv", "details.arxiv_snapshot"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agent%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B202610060000+TO+202610200000%5D&start=0&max_results=200&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T08:27:26.236797+00:00", "role": "primary", "sha256": "80bed80786f939973d0d13b4f892300fee02a6110e326247131fc83d257796de", "supports": ["name", "kind", "scope"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agent%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22+OR+ti%3A%22browser+agents%22+OR+ti%3A%22mobile+agent%22+OR+ti%3A%22mobile+agents%22+OR+ti%3A%22desktop+agent%22%29+AND+submittedDate%3A%5B202610130000+TO+202610192359%5D&start=0&max_results=200&sortBy=submittedDate&sortOrder=descending"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T09:17:33.132448+00:00", "role": "primary", "sha256": "daceefecb2e6b910a6c7f2ffe35e89c7070f6282f4f35ee89e885a2e77aa94c4", "supports": ["summary", "environments", "topics", "scope", "notes"], "url": "https://arxiv.org/html/2610.24277v1"}], "id": "arxiv:2610.24277", "identifiers": {"arxiv": "2610.24277"}, "kind": "paper", "last_verified_at": "2026-10-20T08:27:26.236797+00:00", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://arxiv.org/abs/2610.24277"}], "name": "SPIDES: Information-Flow Control for Browser Agents", "notes": "Reviewed the complete abstract and selected v1 threat-model, policy, evaluation and limitations sections. Guarantees assume correct labels/trusted origins and a trusted browser/harness mediating actions; evaluation uses idealized approvals. Origin-level reader sets can miss recipient distinctions within one origin. No unrestricted prompt-injection immunity or independent reproduction is claimed.", "published_at": "2026-10-16", "relations": [], "review_status": "curated", "scope": "core", "summary": "SPIDES propagates HTML integrity and confidentiality labels into agent context and mediates browser actions with information-flow policies and approval gates, evaluated in the labeled MicroNet ecosystem.", "topics": ["safety-security", "web-navigation", "human-in-the-loop", "evaluation"], "updated_at": "2026-10-16", "url": "https://arxiv.org/abs/2610.24277"}
{"aliases": ["AdvWeb"], "authors": [], "details": {"availability": "Public research source and workflow documentation; no standalone trained checkpoint was verified.", "repository_revision": "e01316e33d11ab145daa7cc7ec12d6ddb0bf38a0"}, "environments": ["web"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-17T09:18:18Z", "role": "primary", "sha256": "019d45dd4c383f919f3ac2bf84c672500d63751a49de2dd2bac9c018ffc99881", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "notes", "relations", "details"], "url": "https://raw.githubusercontent.com/AI-secure/AdvAgent/HEAD/README.md"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T09:18:17Z", "role": "primary", "sha256": "3ea44a202a677c0beb744ab05bd687da93a83365b7eb0b7999c3a4c22d7e2608", "supports": ["identity"], "url": "https://api.github.com/repos/AI-secure/AdvAgent"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T09:18:33Z", "role": "primary", "sha256": "95929797eb0717ce26b26ccbdc752bfda18fc0070975b32f09827b4e9fec6ecf", "supports": ["release_status", "licenses.code"], "url": "https://api.github.com/repos/AI-secure/AdvAgent/git/trees/master?recursive=1"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T09:17:48Z", "role": "primary", "sha256": "3ea44a202a677c0beb744ab05bd687da93a83365b7eb0b7999c3a4c22d7e2608", "supports": ["identity"], "url": "https://api.github.com/repos/AI-secure/AdvWeb"}], "id": "github:ai-secure/advagent", "identifiers": {}, "kind": "framework", "last_verified_at": "2026-10-17T09:18:18Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/AI-secure/AdvAgent"}], "name": "AdvAgent (formerly AdvWeb)", "notes": "The official GitHub API resolves AI-secure/AdvWeb to AI-secure/AdvAgent. Public data-generation, SFT/DPO training and evaluation scripts are present, with upstream data and external model APIs required. No repository license was found, and no permission for third-party data or model weights is inferred. The README citation block is empty, so no paper identifier is bound from title similarity alone.", "published_at": null, "relations": [], "review_status": "curated", "scope": "core", "summary": "Research code for constructing and evaluating adversarial inputs to web agents, with a black-box red-team pipeline using Mind2Web data and SeeAct.", "topics": ["safety-security", "robustness", "evaluation"], "updated_at": null, "url": "https://github.com/AI-secure/AdvAgent"}
{"aliases": [], "authors": [], "details": {}, "environments": ["web"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:40:54Z", "role": "primary", "sha256": "0759e1212025ac487c1646877f97526d35db1a62337192800acaff7684471087", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/ChenWu98/agent-attack/HEAD/README.md"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:31:32Z", "role": "primary", "sha256": "b9a83994d8ac36ccdb606a5219511f8bb35ad3948ade4f9f3e28d386221aedfb", "supports": ["relations"], "url": "https://export.arxiv.org/api/query?id_list=2406.12814%2C2406.13719%2C2406.14056%2C2406.14250%2C2406.14314%2C2406.18082%2C2406.19263%2C2407.00993%2C2407.01476%2C2407.01489%2C2407.01511%2C2407.01558%2C2407.03037%2C2407.03913%2C2407.04346%2C2407.05291%2C2407.07061%2C2407.08850%2C2407.09018%2C2407.09295%2C2407.10956%2C2407.13032%2C2407.13063%2C2407.15673%2C2407.15711%2C2407.16741%2C2407.17490%2C2407.18901%2C2407.19056%2C2407.20183%2C2408.00203%2C2408.02544%2C2408.06327%2C2408.07199%2C2408.11824%2C2408.15978%2C2409.02453%2C2409.04566%2C2409.07429%2C2409.08264%2C2409.09354%2C2409.10741%2C2409.11295%2C2409.13711%2C2409.14337%2C2409.14818%2C2409.15441%2C2409.15637%2C2409.17140%2C2409.20566&max_results=50"}], "id": "github:chenwu98/agent-attack", "identifiers": {}, "kind": "benchmark", "last_verified_at": "2026-10-16T06:40:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/ChenWu98/agent-attack"}], "name": "Adversarial Web Agents", "notes": "", "published_at": null, "relations": [{"evidence_url": "https://export.arxiv.org/api/query?id_list=2406.12814%2C2406.13719%2C2406.14056%2C2406.14250%2C2406.14314%2C2406.18082%2C2406.19263%2C2407.00993%2C2407.01476%2C2407.01489%2C2407.01511%2C2407.01558%2C2407.03037%2C2407.03913%2C2407.04346%2C2407.05291%2C2407.07061%2C2407.08850%2C2407.09018%2C2407.09295%2C2407.10956%2C2407.13032%2C2407.13063%2C2407.15673%2C2407.15711%2C2407.16741%2C2407.17490%2C2407.18901%2C2407.19056%2C2407.20183%2C2408.00203%2C2408.02544%2C2408.06327%2C2408.07199%2C2408.11824%2C2408.15978%2C2409.02453%2C2409.04566%2C2409.07429%2C2409.08264%2C2409.09354%2C2409.10741%2C2409.11295%2C2409.13711%2C2409.14337%2C2409.14818%2C2409.15441%2C2409.15637%2C2409.17140%2C2409.20566&max_results=50", "target_id": "arxiv:2406.12814", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Evaluation resources for adversarial attacks on multimodal web agents.", "topics": ["safety-security", "robustness"], "updated_at": null, "url": "https://github.com/ChenWu98/agent-attack"}
{"aliases": [], "authors": [], "details": {}, "environments": ["web"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-16T10:42:34Z", "role": "primary", "sha256": "9c938f8f3f3ee04f459362a8b7d399588fd030ce476cda111e5f23fbf1e31f10", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/facebookresearch/ai-agent-privacy/HEAD/README.md"}], "id": "github:facebookresearch/ai-agent-privacy", "identifiers": {}, "kind": "benchmark", "last_verified_at": "2026-10-16T10:42:34Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/facebookresearch/ai-agent-privacy"}], "name": "AgentDAM", "notes": "", "published_at": null, "relations": [{"evidence_url": "https://raw.githubusercontent.com/facebookresearch/ai-agent-privacy/HEAD/README.md", "target_id": "arxiv:2503.09780", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Web-agent benchmark for measuring unnecessary use of sensitive personal information during end-to-end task execution.", "topics": ["safety-security", "evaluation"], "updated_at": null, "url": "https://github.com/facebookresearch/ai-agent-privacy"}
{"aliases": [], "authors": [], "details": {"benchmark_protocol": {"actions": ["tap/long-tap/swipe", "navigation buttons and text editing", "refuse, ask-consent and complete", "task-specific shortcut actions"], "counts": [], "evaluation": "Appium-backed task evaluators track goal achievement and harm prevention as separate booleans, plus risk-detection step and completion progress. The base evaluator defaults to max_step=15; task-specific overrides must be retained. Results are not a single generic success-rate metric.", "evidence_urls": ["https://raw.githubusercontent.com/jylee425/mobilesafetybench/HEAD/README.md", "https://api.github.com/repos/jylee425/mobilesafetybench/git/trees/release/ver.3?recursive=1", "https://raw.githubusercontent.com/jylee425/mobilesafetybench/HEAD/mobile_safety/environment.py", "https://raw.githubusercontent.com/jylee425/mobilesafetybench/HEAD/mobile_safety/component/parser.py", "https://raw.githubusercontent.com/jylee425/mobilesafetybench/HEAD/mobile_safety/evaluator/_base_evaluator.py"], "focus": ["mobile task helpfulness", "harm prevention during Android interaction"], "notes": "Inspected release/ver.3 branch exposes task definitions, Android environment and per-task evaluators. It requires Android API 34, Appium/ADB and documented third-party app versions; those APKs/accounts were not acquired. The action parser includes semantic shortcuts in addition to screen gestures, so this is not a strictly coordinate-only action setting. No corpus task total is inferred from evaluator-module or file counts. Apache-2.0 applies to repository code, not automatically app/data assets.", "observations": ["screenshot pixels", "parsed UI hierarchy text", "task instruction and interaction history"], "setting": "interactive"}}, "environments": ["android"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:35:58Z", "role": "primary", "sha256": "e733115d9b2fbe910cb3891162f04a89af8cbda25255c8f0bf60f78101e4309a", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations", "details.benchmark_protocol"], "url": "https://raw.githubusercontent.com/jylee425/mobilesafetybench/HEAD/README.md"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T14:14:05Z", "role": "primary", "sha256": "8dc66f6c5087a32c5c9fb7b3104ea6f9e0082255e2ba3cf9ba8244ce5a24da6f", "supports": ["details.benchmark_protocol"], "url": "https://api.github.com/repos/jylee425/mobilesafetybench/git/trees/release/ver.3?recursive=1"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T14:21:09Z", "role": "primary", "sha256": "f56d6345546a9c67a2113cfba5b7148f5bd6b48fac5c34f1a5f4080e2d81ae2f", "supports": ["details.benchmark_protocol"], "url": "https://raw.githubusercontent.com/jylee425/mobilesafetybench/HEAD/mobile_safety/environment.py"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T14:21:10Z", "role": "primary", "sha256": "120487c3a6492371321dd3eebcd65dc59de6a82454d622d7fed3a7d86b8fa6c5", "supports": ["details.benchmark_protocol"], "url": "https://raw.githubusercontent.com/jylee425/mobilesafetybench/HEAD/mobile_safety/component/parser.py"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T14:23:14Z", "role": "primary", "sha256": "918e3aa64ab3f578fb54580da123c1ff4508022b64ce74f88ce28c7a612756a4", "supports": ["details.benchmark_protocol"], "url": "https://raw.githubusercontent.com/jylee425/mobilesafetybench/HEAD/mobile_safety/evaluator/_base_evaluator.py"}, {"method": "license_header", "retrieved_at": "2026-10-16T06:41:58Z", "role": "primary", "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/jylee425/mobilesafetybench/HEAD/LICENSE.txt"}], "id": "github:jylee425/mobilesafetybench", "identifiers": {}, "kind": "benchmark", "last_verified_at": "2026-10-16T06:35:58Z", "licenses": {"code": {"basis": "repository LICENSE header; not a weights/data license", "declared": "Apache-2.0", "retrieved_at": "2026-10-16T06:41:58Z", "url": "https://raw.githubusercontent.com/jylee425/mobilesafetybench/HEAD/LICENSE.txt"}, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/jylee425/mobilesafetybench"}], "name": "MobileSafetyBench", "notes": "", "published_at": null, "relations": [], "review_status": "curated", "scope": "core", "summary": "Android emulator testbed that scores mobile device-control agents on both safety and helpfulness, using ADB and Appium for observations, actions and success checks.", "topics": ["safety-security", "evaluation"], "updated_at": null, "url": "https://github.com/jylee425/mobilesafetybench"}
{"aliases": [], "authors": [], "details": {"benchmark_protocol": {"actions": ["web interactions"], "counts": [{"basis": "reported", "evidence_url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending", "population": "safe split", "qualifier": "exact", "reference_url": "https://arxiv.org/abs/2503.04957", "unit": "tasks", "value": 250}, {"basis": "reported", "evidence_url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending", "population": "harmful split", "qualifier": "exact", "reference_url": "https://arxiv.org/abs/2503.04957", "unit": "tasks", "value": 250}], "evaluation": "Task outcomes and the Agent Risk Assessment categories evaluate agent behavior on benign and harmful requests.", "evidence_urls": ["https://raw.githubusercontent.com/McGill-NLP/safearena/HEAD/README.md", "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"], "focus": ["web-agent safety", "task completion"], "notes": "Use SafeArena-specific containers rather than ordinary WebArena instances. The documented task splits require access approval; the profile does not imply unrestricted dataset access.", "observations": [], "setting": "interactive"}, "leaderboard_url": "https://huggingface.co/spaces/McGill-NLP/safearena-leaderboard"}, "environments": ["web"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-16T09:36:33Z", "role": "primary", "sha256": "ae50121aa60ac26b32d8d6ad6f578f9091a81a3cb9ca28b98050bddee16c929a", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations", "details.benchmark_protocol", "details.leaderboard_url"], "url": "https://raw.githubusercontent.com/McGill-NLP/safearena/HEAD/README.md"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:54Z", "role": "primary", "sha256": "4953a8e0f029016aef668bce98cec4d2873a7ae49c1fe33c4311fa5eb8d24089", "supports": ["details.benchmark_protocol"], "url": "https://export.arxiv.org/api/query?search_query=%28ti%3A%22computer+use%22+OR+ti%3A%22computer-use%22+OR+ti%3A%22GUI+agent%22+OR+ti%3A%22GUI+agents%22+OR+ti%3A%22GUI+grounding%22+OR+ti%3A%22web+agents%22+OR+ti%3A%22browser+agent%22%29+AND+submittedDate%3A%5B200001010000+TO+202610160636%5D&start=700&max_results=100&sortBy=submittedDate&sortOrder=descending"}], "id": "github:mcgill-nlp/safearena", "identifiers": {}, "kind": "benchmark", "last_verified_at": "2026-10-16T09:36:33Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/McGill-NLP/safearena"}], "name": "SafeArena", "notes": "", "published_at": null, "relations": [{"evidence_url": "https://raw.githubusercontent.com/McGill-NLP/safearena/HEAD/README.md", "target_id": "arxiv:2503.04957", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Web-agent benchmark pairing benign and harmful tasks to study deliberate misuse risks.", "topics": ["safety-security", "evaluation"], "updated_at": null, "url": "https://github.com/McGill-NLP/safearena"}
{"aliases": [], "authors": [], "details": {}, "environments": ["desktop"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:40:54Z", "role": "primary", "sha256": "42bf4aea8369ac2909bf27bcb1078756c2c26567e2b46284c8f3a9e4c4f636c1", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/OS-Copilot/OS-Sentinel/HEAD/README.md"}, {"method": "license_header", "retrieved_at": "2026-10-16T06:42:07Z", "role": "primary", "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/OS-Copilot/OS-Sentinel/HEAD/LICENSE"}], "id": "github:os-copilot/os-sentinel", "identifiers": {}, "kind": "framework", "last_verified_at": "2026-10-16T06:40:54Z", "licenses": {"code": {"basis": "repository LICENSE header; not a weights/data license", "declared": "Apache-2.0", "retrieved_at": "2026-10-16T06:42:07Z", "url": "https://raw.githubusercontent.com/OS-Copilot/OS-Sentinel/HEAD/LICENSE"}, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/OS-Copilot/OS-Sentinel"}], "name": "OS-Sentinel", "notes": "", "published_at": null, "relations": [{"evidence_url": "https://raw.githubusercontent.com/OS-Copilot/OS-Sentinel/HEAD/README.md", "target_id": "arxiv:2510.24411", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Safety-checking environment and benchmark for mobile agents on AndroidWorld that judges proposed actions with rule-based and VLM-based checks, evaluated on the MobileRisk trajectory dataset.", "topics": ["safety-security"], "updated_at": null, "url": "https://github.com/OS-Copilot/OS-Sentinel"}
{"aliases": [], "authors": [], "details": {}, "environments": ["web"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:59:54Z", "role": "primary", "sha256": "61eb6109404f266fd22c6fa8ca58ef773a9d502567403f42edc096958ab2e430", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/OSU-NLP-Group/EIA_against_webagent/HEAD/README.md"}, {"method": "license_header", "retrieved_at": "2026-10-18T11:03:46Z", "role": "primary", "sha256": "953ff47dbbc0c0563d1ec2d245b06d1206f2f38c4ed2f0da728e027ae2e5a973", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/OSU-NLP-Group/EIA_against_webagent/HEAD/LICENSE"}], "id": "github:osu-nlp-group/eia_against_webagent", "identifiers": {}, "kind": "benchmark", "last_verified_at": "2026-10-18T10:59:54Z", "licenses": {"code": {"basis": "repository LICENSE header; not a weights/data license", "declared": "MIT", "retrieved_at": "2026-10-18T11:03:46Z", "url": "https://raw.githubusercontent.com/OSU-NLP-Group/EIA_against_webagent/HEAD/LICENSE"}, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/OSU-NLP-Group/EIA_against_webagent"}], "name": "EIA", "notes": "", "published_at": null, "relations": [{"evidence_url": "https://raw.githubusercontent.com/OSU-NLP-Group/EIA_against_webagent/HEAD/README.md", "target_id": "arxiv:2409.11295", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Evaluation suite for Environmental Injection Attacks that embed invisible malicious HTML elements into benign websites to test privacy leakage in generalist web agents.", "topics": ["safety-security", "evaluation", "web-navigation"], "updated_at": null, "url": "https://github.com/OSU-NLP-Group/EIA_against_webagent"}
{"aliases": [], "authors": [], "details": {}, "environments": ["web", "desktop"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-17T02:44:04Z", "role": "primary", "sha256": "782dea8635b417bc89d21489cacea215ee6878ba5a1d4b9e18ca7b2913545ab6", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/OSU-NLP-Group/RedTeamCUA/HEAD/README.md"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T02:44:04Z", "role": "primary", "sha256": "10a9bb268e611da40ae6edba6777917abb6e97d43dcb480e772ae8169b687e10", "supports": ["identity"], "url": "https://api.github.com/repos/OSU-NLP-Group/RedTeamCUA"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T02:44:04Z", "role": "primary", "sha256": "712f1b7bc13b593cf23ecfab72cec33d1c256afe23a383e49d7d5b2996a0f55c", "supports": ["release_status"], "url": "https://api.github.com/repos/OSU-NLP-Group/RedTeamCUA/git/trees/main?recursive=1"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T02:45:20Z", "role": "primary", "sha256": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/OSU-NLP-Group/RedTeamCUA/HEAD/LICENSE"}, {"method": "license_header", "retrieved_at": "2026-10-17T02:45:20Z", "role": "primary", "sha256": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/OSU-NLP-Group/RedTeamCUA/HEAD/LICENSE"}], "id": "github:osu-nlp-group/redteamcua", "identifiers": {}, "kind": "framework", "last_verified_at": "2026-10-17T02:44:04Z", "licenses": {"code": {"basis": "repository LICENSE header; not a weights/data license", "declared": "Apache-2.0", "retrieved_at": "2026-10-17T02:45:20Z", "url": "https://raw.githubusercontent.com/OSU-NLP-Group/RedTeamCUA/HEAD/LICENSE"}, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/OSU-NLP-Group/RedTeamCUA"}], "name": "RedTeamCUA", "notes": "The release includes environment code, adversarial scenario configuration and RTC-Bench task-generation assets. The project declares Apache-2.0, while incorporated environments retain their licenses. It distinguishes full task runs from tests initialized at injection locations; those settings must not be treated as interchangeable results.", "published_at": null, "relations": [{"evidence_url": "https://raw.githubusercontent.com/OSU-NLP-Group/RedTeamCUA/HEAD/README.md", "target_id": "arxiv:2505.21936", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "A controlled testing framework that combines desktop virtual machines and replicated websites to evaluate computer-use agents against indirect prompt injection.", "topics": ["safety-security", "evaluation", "sandbox"], "updated_at": null, "url": "https://github.com/OSU-NLP-Group/RedTeamCUA"}
{"aliases": [], "authors": [], "details": {}, "environments": ["web"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:59:53Z", "role": "primary", "sha256": "be53561dcf69cb2295a26eaebf965eead10b3504fca30f71f73dc5e5728fc5b3", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/OSU-NLP-Group/WebGuard/HEAD/README.md"}, {"method": "license_header", "retrieved_at": "2026-10-18T11:03:46Z", "role": "primary", "sha256": "93b43d692b033b76129504448695bfe76ef22d18b11e51352bfab4b5622e5aaa", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/OSU-NLP-Group/WebGuard/HEAD/LICENSE"}], "id": "github:osu-nlp-group/webguard", "identifiers": {}, "kind": "benchmark", "last_verified_at": "2026-10-18T10:59:53Z", "licenses": {"code": {"basis": "repository LICENSE header; not a weights/data license", "declared": "MIT", "retrieved_at": "2026-10-18T11:03:46Z", "url": "https://raw.githubusercontent.com/OSU-NLP-Group/WebGuard/HEAD/LICENSE"}, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/OSU-NLP-Group/WebGuard"}], "name": "WebGuard", "notes": "", "published_at": null, "relations": [{"evidence_url": "https://raw.githubusercontent.com/OSU-NLP-Group/WebGuard/HEAD/README.md", "target_id": "arxiv:2507.14293", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Codebase, annotation toolkit, and action-level risk dataset for training and evaluating guardrail models that flag unsafe browser-agent actions before execution.", "topics": ["safety-security", "evaluation", "web-navigation"], "updated_at": null, "url": "https://github.com/OSU-NLP-Group/WebGuard"}
{"aliases": [], "authors": [], "details": {}, "environments": ["web"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:36:05Z", "role": "primary", "sha256": "8682f83d71dd4d44772db3e531f2a9fcb168729211e0cc43493020a0773b0aba", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/pixiebrix/agent-browser-shield/HEAD/README.md"}], "id": "github:pixiebrix/agent-browser-shield", "identifiers": {}, "kind": "tool", "last_verified_at": "2026-10-16T06:36:05Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/pixiebrix/agent-browser-shield"}], "name": "Agent Browser Shield", "notes": "", "published_at": null, "relations": [], "review_status": "curated", "scope": "core", "summary": "Browser extension for filtering potentially unsafe agent input.", "topics": ["safety-security"], "updated_at": null, "url": "https://github.com/pixiebrix/agent-browser-shield"}
{"aliases": [], "authors": [], "details": {}, "environments": ["web"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-16T06:40:54Z", "role": "primary", "sha256": "24251854f9481af5c8db65750197604546f000b54868930a6de79adc365db854", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/SALT-NLP/PopupAttack/HEAD/README.md"}], "id": "github:salt-nlp/popupattack", "identifiers": {}, "kind": "benchmark", "last_verified_at": "2026-10-16T06:40:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/SALT-NLP/PopupAttack"}], "name": "PopupAttack", "notes": "", "published_at": null, "relations": [], "review_status": "curated", "scope": "core", "summary": "Resources for evaluating GUI agents under malicious popup instructions.", "topics": ["safety-security"], "updated_at": null, "url": "https://github.com/SALT-NLP/PopupAttack"}
{"aliases": [], "authors": [], "details": {"license_review": {"code": {"evidence_urls": ["https://raw.githubusercontent.com/scaleapi/browser-art/HEAD/README.md", "https://raw.githubusercontent.com/scaleapi/browser-art/HEAD/LICENSE"], "reason": "The retained README states that the project is MIT licensed, while the retained LICENSE contains Creative Commons Attribution-NonCommercial-NoDerivatives 4.0. These declarations conflict, so a settled code license is not asserted."}}}, "environments": ["web"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:59:54Z", "role": "primary", "sha256": "a9d33536fae38485eb6aac9698ae22ed3533510d43f783d6c0e9f9cf95d9c58f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations", "licenses.code", "notes"], "url": "https://raw.githubusercontent.com/scaleapi/browser-art/HEAD/README.md"}, {"method": "license_header", "retrieved_at": "2026-10-18T11:03:46Z", "role": "primary", "sha256": "f7f28b8c7a1af76b9874ca6d040e8b1eb6768fd1043106c9d315090ea96754e7", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/scaleapi/browser-art/HEAD/LICENSE"}, {"method": "license_header", "retrieved_at": "2026-10-18T11:03:46Z", "role": "primary", "sha256": "f7f28b8c7a1af76b9874ca6d040e8b1eb6768fd1043106c9d315090ea96754e7", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/scaleapi/browser-art/HEAD/LICENSE"}], "id": "github:scaleapi/browser-art", "identifiers": {}, "kind": "benchmark", "last_verified_at": "2026-10-18T10:59:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/scaleapi/browser-art"}], "name": "BrowserART", "notes": "The retained README and LICENSE contain conflicting code-license declarations. The catalog leaves the code-license value unknown pending an upstream clarification; it does not infer data or model permissions.", "published_at": null, "relations": [{"evidence_url": "https://raw.githubusercontent.com/scaleapi/browser-art/HEAD/README.md", "target_id": "arxiv:2410.13886", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Red-teaming benchmark and test suite from Scale AI evaluating whether refusal-trained language models can be jailbroken when deployed as browser agents.", "topics": ["safety-security", "evaluation", "web-navigation"], "updated_at": null, "url": "https://github.com/scaleapi/browser-art"}
{"aliases": [], "authors": [], "details": {}, "environments": ["web"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:59:54Z", "role": "primary", "sha256": "e712feb15886d5a2aba1051d419dc69dc6e58c9a86a9d2d566090bbbf67492fc", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/segev-shlomov/ST-WebAgentBench/HEAD/README.md"}], "id": "github:segev-shlomov/st-webagentbench", "identifiers": {}, "kind": "benchmark", "last_verified_at": "2026-10-18T10:59:54Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/segev-shlomov/ST-WebAgentBench"}], "name": "ST-WebAgentBench", "notes": "", "published_at": null, "relations": [{"evidence_url": "https://raw.githubusercontent.com/segev-shlomov/ST-WebAgentBench/HEAD/README.md", "target_id": "arxiv:2410.06703", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "BrowserGym-based benchmark for evaluating the safety and policy compliance of enterprise web agents across six trustworthiness risk dimensions.", "topics": ["safety-security", "evaluation", "web-navigation"], "updated_at": null, "url": "https://github.com/segev-shlomov/ST-WebAgentBench"}
{"aliases": [], "authors": [], "details": {}, "environments": ["web", "android", "cross-platform"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:59:55Z", "role": "primary", "sha256": "d69da19edaed185ad3071a97db3bb3fcfbdb14c0b514b7d27794b39fbe8e2541", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/thu-ml/MLA-Trust/HEAD/README.md"}, {"method": "license_header", "retrieved_at": "2026-10-18T11:03:46Z", "role": "primary", "sha256": "240fb6820e7bcb0753617bcad6ffbd4a9ba83ca98e1e6aaa06c7b197b244bf16", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/thu-ml/MLA-Trust/HEAD/LICENSE"}], "id": "github:thu-ml/mla-trust", "identifiers": {}, "kind": "benchmark", "last_verified_at": "2026-10-18T10:59:55Z", "licenses": {"code": {"basis": "repository LICENSE header; not a weights/data license", "declared": "MIT", "retrieved_at": "2026-10-18T11:03:46Z", "url": "https://raw.githubusercontent.com/thu-ml/MLA-Trust/HEAD/LICENSE"}, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/thu-ml/MLA-Trust"}], "name": "MLA-Trust", "notes": "", "published_at": null, "relations": [{"evidence_url": "https://raw.githubusercontent.com/thu-ml/MLA-Trust/HEAD/README.md", "target_id": "arxiv:2506.01616", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Unified evaluation framework measuring truthfulness, controllability, safety, and privacy of multimodal LLM agents across interactive web and mobile GUI tasks.", "topics": ["safety-security", "evaluation", "robustness"], "updated_at": null, "url": "https://github.com/thu-ml/MLA-Trust"}
{"aliases": [], "authors": [], "details": {"benchmark_protocol": {"actions": ["OSWorld/PyAutoGUI desktop interaction"], "counts": [{"basis": "reported", "evidence_url": "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/README.md", "population": "reported benchmark total; not a recount of unique released task IDs", "qualifier": "exact", "unit": "tasks as reported by the paper/README", "value": 150}, {"basis": "computed", "evidence_url": "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/evaluation_examples/test_misuse.json", "population": "released misuse task-ID list; before injection/goal expansion", "qualifier": "exact", "unit": "task IDs", "value": 50}, {"basis": "computed", "evidence_url": "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/evaluation_examples/test_injection.json", "population": "released injection task-ID list; before injection/goal expansion", "qualifier": "exact", "unit": "task IDs", "value": 10}, {"basis": "computed", "evidence_url": "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/evaluation_examples/test_misbehavior.json", "population": "released misbehavior task-ID list; before injection/goal expansion", "qualifier": "exact", "unit": "task IDs", "value": 50}], "evaluation": "Automated LLM judge scores both task accuracy and safety from saved execution traces. Keep judge model/prompt version and harm category with a result; the scripts separately compute predicted metrics. This is an interactive desktop safety benchmark, not a single offline grounding score.", "evidence_urls": ["https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/README.md", "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/run.py", "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/evaluation_examples/test_misuse.json", "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/evaluation_examples/test_injection.json", "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/evaluation_examples/test_misbehavior.json"], "focus": ["deliberate misuse", "prompt-injection safety", "unintended harmful behavior during normal tasks"], "notes": "README/paper reports 150 tasks, whereas retained misuse/injection/misbehavior lists contain 50/10/50 IDs. run.py expands injection vectors/goals, so the released ID inventory is not a direct recount of paper settings; the complete expanded total was not reconstructed. Judge agreement F1 values in the README measure agreement with human labels, not agent success. Experiments use OSWorld desktop apps, with VMware documented as tested and screenshot/a11y/SOM observation choices. Code Apache-2.0 does not establish VM/app/data rights. Publisher asks that the benchmark not be used directly as agent training data.", "observations": ["screenshot", "accessibility tree", "screenshot plus accessibility tree", "set-of-marks"], "setting": "interactive"}, "evaluation_type": "automated judgment of accuracy and safety", "metrics": ["task completion", "safety"], "task_count": 150, "task_unit": "tasks reported by the paper/README; not a recount of released task IDs"}, "environments": ["desktop"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-16T09:36:33Z", "role": "primary", "sha256": "d001158f58ac37b01c8470224db1567a52feb8010877305f7966637cf7400706", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations", "details", "details.benchmark_protocol", "notes"], "url": "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/README.md"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T14:23:14Z", "role": "primary", "sha256": "86dd59a252fd654fc3244268a75de37b2c7f97e16f6b1de8c4cf885458a422f7", "supports": ["details.benchmark_protocol", "notes"], "url": "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/run.py"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T14:22:04Z", "role": "primary", "sha256": "8ff6092a51cc924c5225ace01c011c6481a2461eb0a47b7360eddaf6619b898c", "supports": ["details.benchmark_protocol", "notes"], "url": "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/evaluation_examples/test_misuse.json"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T14:22:04Z", "role": "primary", "sha256": "b4c451236bb4ff36eb7e790dfa8cb517bffb245602bc441080314b13c8aba54f", "supports": ["details.benchmark_protocol", "notes"], "url": "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/evaluation_examples/test_injection.json"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T14:22:05Z", "role": "primary", "sha256": "8eb6bade72696c3c0b53160a05b006a773c9e553ab5e401418b54d9acc98c0f2", "supports": ["details.benchmark_protocol", "notes"], "url": "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/evaluation_examples/test_misbehavior.json"}, {"method": "license_header", "retrieved_at": "2026-10-16T09:39:34Z", "role": "primary", "sha256": "7c3ab1eeeb073d92238ecb65ffaebe58ab5922929b115a7c0b64eb4f9a9afc2f", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/LICENSE"}], "id": "github:tml-epfl/os-harm", "identifiers": {}, "kind": "benchmark", "last_verified_at": "2026-10-16T09:36:33Z", "licenses": {"code": {"basis": "repository LICENSE header; not a weights/data license", "declared": "Apache-2.0", "retrieved_at": "2026-10-16T09:39:34Z", "url": "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/LICENSE"}, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/tml-epfl/os-harm"}], "name": "OS-Harm", "notes": "The retained task-ID manifests have 50 misuse, 10 injection and 50 misbehavior IDs. Prompt-injection execution expands vectors/goals; the reported 150-task total must not be read as 150 unique released IDs.", "published_at": null, "relations": [{"evidence_url": "https://raw.githubusercontent.com/tml-epfl/os-harm/HEAD/README.md", "target_id": "arxiv:2506.14866", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Benchmark for safety of computer-use agents across misuse, prompt injection, and unintended model behavior.", "topics": ["safety-security", "evaluation"], "updated_at": null, "url": "https://github.com/tml-epfl/os-harm"}
{"aliases": [], "authors": [], "details": {}, "environments": ["android"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:59:55Z", "role": "primary", "sha256": "2403ef3ec300e257dc631e3c569fc9aabaab1d3f385ac37abe1e4021aaae6e21", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/VeriSafeAgent/VeriSafeAgent/HEAD/README.md"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-18T13:06:20.028238+00:00", "role": "primary", "sha256": "9b04efe4136a39e3e5f21a1d3bb8a44075f1370bdbeec6a4307783f3836f8311", "supports": ["summary", "relations", "notes"], "url": "https://arxiv.org/html/2503.18492v2"}], "id": "github:verisafeagent/verisafeagent", "identifiers": {}, "kind": "framework", "last_verified_at": "2026-10-18T10:59:55Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/VeriSafeAgent/VeriSafeAgent"}], "name": "VeriSafe Agent", "notes": "The verification method requires developers to supply app-state predicates and transition hooks. The paper-to-repository relation is supported by the exact-version paper, not inferred from the repository name.", "published_at": null, "relations": [{"evidence_url": "https://arxiv.org/html/2503.18492v2", "target_id": "arxiv:2503.18492", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Runtime verification framework for Android GUI agents that checks proposed actions against logical rules derived from user instructions and developer-defined app states.", "topics": ["safety-security", "mobile-navigation", "state-tracking"], "updated_at": null, "url": "https://github.com/VeriSafeAgent/VeriSafeAgent"}
{"aliases": [], "authors": [], "details": {"license_review": {"code": {"evidence_urls": ["https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/LICENSE", "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/NOTICE"], "reason": "Record only the publisher's top-level Apache-2.0 declaration for Orbit-original code. NOTICE explicitly preserves per-component terms; neither bundled or external data nor model weights inherit this declaration. Its account of BrowserART licensing is not an independent resolution of upstream conflicting declarations."}}}, "environments": ["web", "desktop"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-20T13:41:22.467183+00:00", "role": "primary", "sha256": "b6743e3e9e5256d7afa8bb411baf246b60ba71bcfb84eda5a816fba3ddd7d78b", "supports": ["name", "kind", "summary", "topics", "scope", "notes"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/README.md"}, {"method": "github_repository_identity", "retrieved_at": "2026-10-20T13:39:57.052202+00:00", "role": "primary", "sha256": "0436414b9bcc4307f3bcde476a544da6fe1dd5baa90939eed392942490fbfbf7", "supports": ["id", "url"], "url": "https://api.github.com/repos/wlanderson0/orbit"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T13:42:37.098639+00:00", "role": "primary", "sha256": "c47d046b129999da5e63b7723a3f3e2e0cc31c27ce74b45f71f8c6d0df1fc0d8", "supports": ["summary", "environments", "scope", "notes"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/orbit/scenarios/browser/browserart/tools.py"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T14:10:32.910533Z", "role": "primary", "sha256": "2cce76b22252a415d64c356b2062771e528b554dd83b6f88fca769f48fa8a826", "supports": ["summary", "environments", "scope", "notes"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/orbit/scenarios/desktop/osworld/tools.py"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T14:09:35.060985Z", "role": "primary", "sha256": "66515d7ee87eab5a2f42c26132d0bf3e2a3b9c006156d2378181cf91966a2903", "supports": ["notes"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/docs/benchmark-qualification.md"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T14:10:29.541892Z", "role": "primary", "sha256": "94c781028dc80b2a748d759fae2456f6a986da7487651a6bcbb179e8ad4fe023", "supports": ["notes"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/orbit/scenarios/browser/browserart/scorer.py"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T13:41:29.036076+00:00", "role": "primary", "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30", "supports": ["licenses.code"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/LICENSE"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T14:09:31.550310Z", "role": "primary", "sha256": "f68a64f775b1f329416f2fe294bb07b5e475893f77d1cdb5300bdb08acf95410", "supports": ["licenses.code", "licenses.data", "notes"], "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/NOTICE"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-20T13:12:28.833839+00:00", "role": "primary", "sha256": "3ea667f4187ba601011e5ff00c4d63676b6d9903d12b80a0d7d5dd9920f00c65", "supports": ["relations"], "url": "https://arxiv.org/html/2609.33102v1"}], "id": "github:wlanderson0/orbit", "identifiers": {}, "kind": "framework", "last_verified_at": "2026-10-20T13:41:22.467183+00:00", "licenses": {"code": {"basis": "Complete root Apache License 2.0 text for Orbit-original code. NOTICE preserves separate third-party component terms, including MIT-vendored modules; this is not a uniform license for the entire dependency/data bundle.", "declared": "Apache-2.0", "retrieved_at": "2026-10-20T13:41:29.036076+00:00", "url": "https://raw.githubusercontent.com/wlanderson0/orbit/588b3035f92450ec4f0496bb8b629daa205e84c7/LICENSE"}, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/wlanderson0/orbit"}], "name": "ORBIT framework", "notes": "Reviewed version 1.0.3 source at commit 588b3035, not an asserted reproduction of the paper. BrowserART tools return accessibility trees, including the screenshot-named refresh. Qualification is scenario-specific: the project documents browser-judge and desktop-scoring limits; its opt-in exact-send oracle covers one derived local task, not SMTP delivery. Runtime profiles and scoring policies are separate. Apache-2.0 is the top-level Orbit code declaration; third-party components and external data/environments retain separate terms. No installation, release archive or live evaluation was tested.", "published_at": null, "relations": [{"evidence_url": "https://arxiv.org/html/2609.33102v1", "target_id": "arxiv:2609.33102", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Inspect-based multi-agent evaluation framework with role, scheduling and communication controls, BrowserART accessibility-tree actions, and desktop mouse, keyboard and screenshot tools.", "topics": ["evaluation", "safety-security", "web-navigation", "observability"], "updated_at": null, "url": "https://github.com/wlanderson0/orbit"}
{"aliases": [], "authors": [], "details": {}, "environments": ["android"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-16T13:14:29Z", "role": "primary", "sha256": "15d40c50df8493575a94aa2aa4913c654fea7b7f821d99b7bf443a64f1cc50d3", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations"], "url": "https://raw.githubusercontent.com/Wuzheng02/OS-Kairos/HEAD/README.md"}], "id": "github:wuzheng02/os-kairos", "identifiers": {}, "kind": "agent", "last_verified_at": "2026-10-16T13:14:29Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/Wuzheng02/OS-Kairos"}], "name": "OS-Kairos", "notes": "The documented evaluation modes distinguish model-simulated human feedback from actual human interaction; preserve that distinction when interpreting results.", "published_at": null, "relations": [{"evidence_url": "https://raw.githubusercontent.com/Wuzheng02/OS-Kairos/HEAD/README.md", "target_id": "arxiv:2503.16465", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Android GUI-agent research implementation that uses confidence-guided interaction to choose between autonomous actions and human assistance.", "topics": ["human-in-the-loop", "safety-security", "mobile-navigation"], "updated_at": null, "url": "https://github.com/Wuzheng02/OS-Kairos"}
{"aliases": [], "authors": [], "details": {}, "environments": ["unspecified"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-18T10:59:55Z", "role": "primary", "sha256": "282883d02391a86c5a4d0cdafdc1d33f10a5073f31eba8c852095601d0aeb939", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations", "notes"], "url": "https://raw.githubusercontent.com/Wuzheng02/VeriOS/HEAD/README.md"}], "id": "github:wuzheng02/verios", "identifiers": {}, "kind": "framework", "last_verified_at": "2026-10-18T10:59:55Z", "licenses": {"code": null, "data": null, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://github.com/Wuzheng02/VeriOS"}], "name": "VeriOS", "notes": "The released scripts evaluate recorded task inputs; the retained README does not establish a live Android execution environment. No runtime behavior or reported results were independently reproduced.", "published_at": null, "relations": [{"evidence_url": "https://raw.githubusercontent.com/Wuzheng02/VeriOS/HEAD/README.md", "target_id": "arxiv:2509.07553", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Research code for proactive human–GUI-agent verification, including offline interaction-loop and dual-agent evaluation scripts plus LLaMA-Factory fine-tuning configurations.", "topics": ["safety-security", "human-in-the-loop", "mobile-navigation"], "updated_at": null, "url": "https://github.com/Wuzheng02/VeriOS"}
{"aliases": [], "authors": [], "details": {"benchmark_protocol": {"actions": ["browser interaction"], "counts": [{"basis": "reported", "evidence_url": "https://huggingface.co/datasets/microsoft/WebTailBench/raw/main/README.md", "population": "main V1/V2 suite (shared identities)", "qualifier": "exact", "unit": "task identities", "value": 609}, {"basis": "reported", "evidence_url": "https://huggingface.co/datasets/microsoft/WebTailBench/raw/main/README.md", "population": "refusal suite", "qualifier": "exact", "unit": "tasks", "value": 111}], "evaluation": "Task-specific precomputed rubrics; refusal tasks are a separate evaluation population.", "evidence_urls": ["https://huggingface.co/datasets/microsoft/WebTailBench/raw/main/README.md"], "focus": ["live web task completion", "unsafe-request refusal"], "notes": "V2 refreshes task strings and rubrics while preserving V1 task IDs. Counts are task definitions, not independently verified runnable website states.", "observations": ["web pages"], "setting": "interactive"}, "disabled": false, "download_url": "https://huggingface.co/datasets/microsoft/WebTailBench/tree/main", "gated": false, "hub_id": "microsoft/WebTailBench", "license_review": {"data": {"evidence_urls": ["https://huggingface.co/datasets/microsoft/WebTailBench/raw/main/README.md"], "reason": "The benchmark's Hub dataset card declares MIT in its metadata header (license: mit)."}}}, "environments": ["web"], "evidence": [{"method": "editorial_primary_review", "retrieved_at": "2026-10-17T02:42:32Z", "role": "primary", "sha256": "c4cfc361d1d47d7855269af4110483363964e23f9c65244a1182830eccccbb7f", "supports": ["name", "kind", "summary", "environments", "topics", "scope", "relations", "details", "licenses.data"], "url": "https://huggingface.co/datasets/microsoft/WebTailBench/raw/main/README.md"}, {"method": "editorial_primary_review", "retrieved_at": "2026-10-17T02:42:32Z", "role": "primary", "sha256": "93469a20d333219dde89a225891a54ee716120d559eb8ac9f69a1a7ee84f57a3", "supports": ["identity", "release_status", "details", "licenses.data"], "url": "https://huggingface.co/api/datasets/microsoft/WebTailBench"}], "id": "url:3b6dfd52f98bfe76bc7f", "identifiers": {}, "kind": "benchmark", "last_verified_at": "2026-10-17T02:42:32Z", "licenses": {"code": null, "data": {"basis": "publisher model/dataset card metadata; third-party assets retain their own terms", "declared": "mit", "retrieved_at": "2026-10-17T02:42:32Z", "url": "https://huggingface.co/datasets/microsoft/WebTailBench/raw/main/README.md"}, "document": null, "weights": null}, "links": [{"rel": "primary", "url": "https://huggingface.co/datasets/microsoft/WebTailBench"}], "name": "WebTailBench", "notes": "Public TSV files contain V1, V2 and refusal task definitions. The card declares MIT for the dataset. V1 and V2 share the same 609 task identities and must not be summed as distinct tasks. The May 2026 refresh supersedes calendar-bound task wording; live-site availability is not established by this metadata review.", "published_at": null, "relations": [{"evidence_url": "https://huggingface.co/datasets/microsoft/WebTailBench/raw/main/README.md", "target_id": "arxiv:2511.19663", "type": "documented_by"}], "review_status": "curated", "scope": "core", "summary": "Live-web task definitions and rubrics for evaluating computer-use agents, with preserved original tasks, a refreshed V2 task suite and a separate refusal set.", "topics": ["evaluation", "web-navigation", "safety-security"], "updated_at": null, "url": "https://huggingface.co/datasets/microsoft/WebTailBench"}