If you discover a security vulnerability in ai-quarantine, please report it responsibly:
- Do NOT open a public issue.
- Email: security@secai-hub.dev (or open a private advisory on GitHub)
- Include: description, reproduction steps, and impact assessment.
| Step | Target |
|---|---|
| Acknowledge receipt | 48 hours |
| Confirm and assign severity | 5 business days |
| Publish fix (critical/high) | 7 calendar days |
| Publish fix (medium/low) | 30 calendar days |
| Public advisory | After fix is released |
We will coordinate disclosure with the reporter. If a fix requires more time, we will communicate the revised timeline and may issue a temporary mitigation advisory.
This project scans AI model artifacts for safety issues. Vulnerabilities in the scanning pipeline itself (bypass, false negative, sandbox escape) are considered critical.
- Pipeline bypass (artifact promoted without completing all stages)
- Scanner evasion (crafted artifact evades detection)
- Sandbox escape (behavioral test sandbox compromised)
- Audit log tampering (chain integrity circumvented)
- Promotion without authentication (SERVICE_TOKEN bypass)
- Denial of service against the scanning pipeline
- Vulnerabilities in upstream scanners (modelscan, fickling, garak) — report these to the respective projects
- Social engineering attacks
- Physical access attacks
| Version | Supported | Notes |
|---|---|---|
| 0.1.x | Yes | Current stable release |
| < 0.1 | No | Pre-release, unsupported |
Users should always run the latest patch release within their supported major.minor version.