Pre-1.0, only the latest tagged release receives security fixes.

Do not file public issues for security problems.

Email a description and reproduction to the maintainer. Expect an acknowledgement within 72 hours and a fix or mitigation plan within 14 days for issues that can be reproduced.

If no response in 14 days, you are free to disclose publicly.

In scope:

• Vulnerabilities in the scanner itself (RCE in scanner code, credential leakage in reports, path traversal in HTML output)
• False negatives where a real LLM vulnerability is missed by an existing scanner
• Auth or credential mishandling in connectors

Out of scope:

• Vulnerabilities in the LLM endpoints being scanned. Report those to the LLM vendor.
• Behavioral findings against scanned targets. Those are the product's purpose.
• Vulnerabilities in litellm, pydantic, or other upstream dependencies. Report to the respective project.

Reports are handled confidentially. Reporter credit in CHANGELOG and release notes if requested.