Skip to content

Security: hamidmatiny/aegis

Security

SECURITY.md

Security Policy

AEGIS is a security product — vulnerabilities in it are higher-severity than in typical software, since they can undermine the defenses it's meant to provide. Please report them responsibly.

Reporting a vulnerability

Do not open a public GitHub issue for security vulnerabilities.

Instead, use GitHub's private vulnerability reporting for this repository. Include:

  • Affected service(s) (gateway, input-defense, agent-gate, etc.)
  • Steps to reproduce, or a minimal PoC
  • Impact (what an attacker could do, and against what deployment: default self-hosted config vs. hardened production config)

We aim to acknowledge valid reports within 5 business days. We will coordinate remediation and disclosure with the reporter when contact details are provided.

The Oracle public demo also publishes this policy at /security.html and machine-readable contact information at /.well-known/security.txt.

Safe harbor

We support good-faith security research conducted under this policy. We will not pursue legal action for research that avoids privacy violations, service disruption, data destruction, and actions outside this scope. If you are unsure whether a test is permitted, report the concern first and wait for guidance.

Scope

In scope: any of the services in this monorepo (gateway, input-defense, output-defense, policy-engine, model-router, agent-gate, audit, redteam, dashboard, sdk/), and the Docker/Helm deployment manifests under deploy/.

Known, already-documented limitation: adaptive-attacker bypass rates against the detection layer are tracked openly in RESULTS.md — that is expected, ongoing research, not something to report as a new finding unless you have a specific reproducible bypass not already covered there. When citing those results externally, use R1 BR and Adapt BR separately (see the methodology note in RESULTS / redteam README), not a blended “overall bypass rate.”

Out of scope: denial-of-service testing, social engineering, physical attacks, third-party services, and accessing, modifying, or deleting data that is not your own.

Supply chain

  • Dependencies are scanned weekly via Dependabot, OSV-Scanner, govulncheck, pip-audit, and npm audit (see .github/workflows/security.yml).

  • Released container images are signed keylessly with cosign via GitHub OIDC and shipped with a CycloneDX SBOM (see .github/workflows/release.yml). Verify with:

    cosign verify \
      --certificate-identity-regexp "https://github.com/hamidmatiny/aegis/.*" \
      --certificate-oidc-issuer https://token.actions.githubusercontent.com \
      ghcr.io/hamidmatiny/aegis-gateway:latest

There aren't any published security advisories