# AdvAgent (formerly AdvWeb)

Research code for constructing and evaluating adversarial inputs to web agents, with a black-box red-team pipeline using Mind2Web data and SeeAct.

- Catalog URL: https://awesomecua.github.io/resources/github-ai-secure-advagent/
- Original source: https://github.com/AI-secure/AdvAgent
- Kind: framework
- Review: curated
- Scope: core
- Environments: web
- Topics: safety-security, robustness, evaluation
- Evidence checked: 2026-10-17T09:18:18Z
- ID: github:ai-secure/advagent

## Limitations

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.

- Record JSON: https://awesomecua.github.io/resources/github-ai-secure-advagent/record.json

## License declarations

- code: not recorded
- data: not recorded
- document: not recorded
- weights: not recorded

## Primary evidence

- https://raw.githubusercontent.com/AI-secure/AdvAgent/HEAD/README.md (retrieved 2026-10-17T09:18:18Z; SHA-256 019d45dd4c383f919f3ac2bf84c672500d63751a49de2dd2bac9c018ffc99881)
- https://api.github.com/repos/AI-secure/AdvAgent (retrieved 2026-10-17T09:18:17Z; SHA-256 3ea44a202a677c0beb744ab05bd687da93a83365b7eb0b7999c3a4c22d7e2608)
- https://api.github.com/repos/AI-secure/AdvAgent/git/trees/master?recursive=1 (retrieved 2026-10-17T09:18:33Z; SHA-256 95929797eb0717ce26b26ccbdc752bfda18fc0070975b32f09827b4e9fec6ecf)
- https://api.github.com/repos/AI-secure/AdvWeb (retrieved 2026-10-17T09:17:48Z; SHA-256 3ea44a202a677c0beb744ab05bd687da93a83365b7eb0b7999c3a4c22d7e2608)