We release patches for security vulnerabilities for the following versions:
| Version | Supported |
|---|---|
| main | ✅ Active development |
| 3.x | ✅ Current stable |
| < 3.0 | ❌ End of life |
Please do not open public GitHub issues for security vulnerabilities.
- Go to Security Advisories
- Click "Report a vulnerability"
- Fill in the details (see below)
If GitHub Security Advisories are not available, contact maintainers through private channels.
Please provide:
- Description — Clear explanation of the vulnerability
- Affected versions — Which versions are impacted
- Reproduction steps — Minimal code or commands to reproduce
- Impact — What an attacker could achieve
- Suggested fix — If you have ideas for mitigation
| Stage | Target |
|---|---|
| Acknowledgment | Within 48 hours |
| Initial assessment | Within 7 days |
| Fix development | Depends on severity |
| Patch release | As soon as possible |
We will keep you informed throughout the process.
When deploying YOLO-Toys:
- The API validates file types and sizes (
MAX_UPLOAD_MB) - Only image files are accepted for inference endpoints
- WebSocket frames are size-limited
- Run behind a reverse proxy (nginx, Caddy) in production
- Set appropriate
ALLOW_ORIGINSfor CORS - Use HTTPS for all public deployments
- Consider rate limiting for public endpoints
- Docker image runs as non-root user
- Multi-stage build minimizes attack surface
- No sensitive data in image layers
- Keep dependencies updated
- Use
pip-auditto check for known vulnerabilities:
pip install pip-audit
pip-audit- YOLO weights are downloaded from Ultralytics on first use
- HuggingFace models are downloaded from HuggingFace Hub
- Ensure network access to trusted sources only
- Set
MAX_CONCURRENCYto prevent resource exhaustion - Monitor memory usage with large models
- Consider container resource limits in production
Security patches are released as patch versions. Subscribe to GitHub releases or watch the repository for updates.
Thank you for helping keep YOLO-Toys secure! 🔒