已审阅并标注「安全与防护」的研究与资产。
条目依据已审阅标签归类，同一主题不表示资源之间存在依赖关系。日期保留来源精度，注释保留原语言。
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.
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.
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.
Evaluates multimodal LLMs on web interface reasoning, visual perturbation robustness, and safety-critical web understanding tasks.
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.
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.
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.
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.
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.
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.
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.
Introduces SAPA-Bench with 7,138 smartphone scenarios to evaluate whether MLLM-powered mobile agents detect and protect sensitive personal information on screen.
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.
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.
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.
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.
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.
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.
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.
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.
Evaluates agent safety across real browser, code, filesystem, shell, and messaging tools using rule-based and model-based assessments.
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.
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.
Studies chain-level risk detection for mobile multimodal agents and a language-model-assisted evaluator, assessing safety interventions together with functional preservation.
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.
Extends OSWorld with tasks covering user misuse, prompt injection, and model misbehavior, with assessment of both task success and safety.
Demonstrates clean-text visual backdoor attacks (VIBMA) that compromise fine-tuned VLM mobile agents through visual triggers while keeping text instructions intact.
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.
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.
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.
Benchmarks truthfulness, controllability, safety, and privacy of multimodal LLM agents across interactive web and mobile GUI tasks.
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.
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.
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.
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.
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.
Outlines an end-to-end threat model and empirical security evaluation across architectural layers of autonomous browser agents.
Analyzes security vulnerabilities across OEM system-level, third-party, and research mobile LLM agents, covering GUI and cross-application attack surfaces.
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.
Systematizes safety and security threats to computer-using agents, with taxonomies of threats and defensive strategies.
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.
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.
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.
Presents a modular security testing framework that plugs into BrowserGym and tool-use benchmarks to evaluate agents under configurable threat models.
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.
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.
Frames GUI-agent trustworthiness as an execution gap and reviews failures, attacks, and defenses across perception, reasoning, and interaction.
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.
Proposes a guardrail agent that reasons over explicit safety policies and checks action trajectories, with a web-agent safety benchmark.
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.
Translates user instructions into formal specifications and applies runtime logic-based checks to proposed mobile GUI actions.
Evaluates whether autonomous web agents minimize use of sensitive personal information when completing tasks that expose unnecessary private data.
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.
Pairs benign and harmful web tasks to assess deliberate misuse risks and categorize agent behavior by risk level.
Analyzes web-agent vulnerabilities at the component level, examining goal placement in prompts, multistep action generation, and observations as sources of increased susceptibility.
Uses step-level confidence estimates to decide when a GUI agent should act autonomously or request human intervention in complex situations.
Examines active environmental injection against mobile agents, testing whether adversarial interface content redirects decisions during task execution.
Proposes computer-using personal agents connected to user-controlled data repositories, discussing privacy, interoperability, and collaborative tasks.
Tests whether adversarial visual pop-ups divert computer agents from intended tasks in desktop and web benchmark environments.
Introduces CASA to assess web agents' responses to cultural and social norms in shopping and forum tasks, including misleading observations.
第 121–180 条，共 213 条资源。
通过主题导出清单获取完整记录和证据。克隆目录后，可将主题记录直接交给标准 RAG 与引用导出工具。
