PolyGuard is a security tool — we hold ourselves to a high standard. If you find a vulnerability in PolyGuard itself, please help us fix it responsibly.

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

Report security issues by emailing: security@quantumlogics.dev

Include as much of the following as possible:

• A description of the vulnerability and its potential impact
• The affected component (e.g. src/inference/api.py, rules_engine/, Docker setup)
• Steps to reproduce or a minimal proof-of-concept
• The vulnerability type and relevant CWE if known (see our taxonomy in data/labels/vulnerability_types.json)
• Any suggested remediation

We will acknowledge your report within 48 hours and aim to release a fix within 14 days for critical issues.

The following are in scope for security reports:

The following are out of scope:

• Vulnerabilities in code submitted to PolyGuard for scanning (that's the point — we're supposed to find those)
• Theoretical attacks with no practical exploit path
• Rate-limiting or abuse on the public HuggingFace Spaces demo

These are acknowledged design decisions, not bugs:

configs/api_config.yaml — secret_key: "change-me-in-production" The JWT secret key placeholder must be replaced before any production deployment. Never run with the default value. See .env.example for the correct environment variable pattern.

api.py — Unauthenticated scan endpoint The public /analyze endpoint intentionally requires no auth for the open demo. Production deployments should sit behind authentication middleware or an API gateway.

Model input trust boundary Code submitted to the ML model is treated as untrusted text. The pipeline does not execute submitted code. However, extremely long inputs may affect inference latency — set MAX_CODE_LENGTH in your environment to enforce limits.

We monitor dependencies via GitHub Dependabot. If you spot a CVE in requirements.txt that Dependabot has not caught, please open a security report as described above.

To audit locally:

pip install pip-audit
pip-audit -r requirements.txt

We follow coordinated disclosure:

1. Reporter notifies us privately.
2. We confirm, reproduce, and develop a fix.
3. We release the fix and credit the reporter (unless anonymity is requested).
4. Details are published in the release notes after users have had time to update.

Researchers who have responsibly disclosed issues will be credited here.

No disclosures yet — be the first!

PolyGuard — Secure code is not optional.