Once Agent OS reaches 1.0, we will maintain security updates for the latest minor version.

Please do not report security vulnerabilities through public GitHub issues.

If you discover a security vulnerability in Agent OS, please report it responsibly:

Send an email to: security@agent-os.dev

Include:

• Description of the vulnerability
• Steps to reproduce
• Potential impact
• Any suggested fixes (optional)

You can also report via GitHub Security Advisories.

For sensitive reports, you may encrypt your message using our PGP key:

-----BEGIN PGP PUBLIC KEY BLOCK-----
[PGP key to be added]
-----END PGP PUBLIC KEY BLOCK-----

Key fingerprint: [To be added]

We use the following severity levels:

Agent OS provides application-level policy enforcement:

• ✅ Deterministic policy checks on agent actions
• ✅ SQL injection prevention (via policy rules)
• ✅ File system access restrictions (configurable paths)
• ✅ Rate limiting and resource controls
• ✅ Action logging and audit trails (Flight Recorder)

Agent OS is middleware, not a security sandbox:

• ❌ Memory corruption or process isolation (use containers)
• ❌ Compromised LLM providers
• ❌ Network-level attacks
• ❌ Supply chain attacks on dependencies

For production deployments, we recommend:

• Running agents in isolated containers
• Using network policies to restrict agent communication
• Monitoring the Flight Recorder for anomalies
• Regular dependency audits

1. Keep Agent OS updated - Always run the latest version
2. Use restrictive policies - Start with minimal permissions, expand as needed
3. Enable Flight Recorder - Log all agent actions for audit
4. Review policies regularly - Ensure policies match current requirements
5. Run in containers - Isolate agents for defense in depth

1. Sign commits - Use git commit -s (DCO) and GPG signing
2. Review dependencies - Check for known vulnerabilities before adding
3. Write secure code - Follow OWASP guidelines
4. Add tests - Include security-relevant test cases
5. Document security implications - Note any security considerations in PRs

We monitor dependencies for known vulnerabilities using:

• GitHub Dependabot
• Regular security audits

We follow coordinated disclosure:

1. Reporter notifies us privately
2. We confirm and assess the vulnerability
3. We develop and test a fix
4. We coordinate disclosure timing with the reporter
5. Fix is released with security advisory
6. Credit is given to the reporter (unless they prefer anonymity)

Published security advisories are available at: github.com/imran-siddique/agent-os/security/advisories

We recognize security researchers who responsibly disclose vulnerabilities:

No submissions yet - be the first!

• Security issues: security@agent-os.dev
• General questions: GitHub Discussions

Last updated: February 2026