ReverseAI analyzes and executes software it does not trust. Please read the full
threat model in docs/SECURITY.md.
Open a private security advisory on this repository. Do not file a public issue with working exploit details.
Please include:
- the affected version or commit,
- the smallest reproduction you can share,
- what you expected versus what happened,
- whether the issue crosses a trust boundary (host, sandbox, API, model), and
- any suggested fix.
The main branch and the latest release receive security fixes. Older releases
may not.
In scope: the API, auth and path guard, sandbox backends, archive handling, SSRF protections, preview isolation, prompt-injection handling, secret handling, and the Docker image.
Out of scope: findings in optional third-party tools (Ghidra, Qiling, Tesseract, Playwright) that are not caused by ReverseAI's integration; and attacks that require the operator to enable a documented unsafe option.
See the "Known limitations" section of docs/SECURITY.md.
We would rather state a gap than imply protection we do not have.