Proxy AI is security software, so vulnerability reports are treated seriously.

Please do not open a public GitHub issue for security vulnerabilities.

Preferred reporting path:

1. Open a private GitHub Security Advisory for this repository.
2. Include a minimal reproduction using synthetic data.
3. Describe impact, affected component, and expected behavior.

If private advisories are not available, contact the maintainers privately before sharing exploit details publicly.

• Real API keys, database credentials, private keys, passwords, PII, or customer data.
• Screenshots containing live tokens or internal hostnames.
• Public proof-of-concept payloads that enable credential exfiltration before a fix is available.

The public project is currently pre-1.0. Security fixes are made on main until formal release branches exist.

In scope:

• Sanitization bypasses that leak raw sensitive values to upstream providers.
• Rehydration defects that expose placeholders or corrupt sensitive values.
• MITM proxy behavior that causes unsafe traffic handling.
• Audit/export paths that expose raw secrets unexpectedly.
• Admin routes that bypass role checks.

Out of scope:

• Reports using real third-party credentials.
• Attacks requiring local administrator access without additional impact.
• Denial-of-service reports without a practical security consequence.