Skip to content

Security: sentinel-safety/SENTINEL

Security

SECURITY.md

✅——————# Security Policy

Supported Versions

Version Supported
1.x ✅ Yes

Security fixes are applied to the latest release on main.


Reporting a Vulnerability

Do not report security vulnerabilities through public GitHub issues.

Report privately to: [email protected]
Subject line: [SECURITY] Brief description

What to include

  • Description of the vulnerability and its potential impact
  • Steps to reproduce or proof-of-concept
  • Which component is affected (API, scoring engine, audit log, federation layer, SDK, etc.)
  • Suggested severity: critical / high / medium / low

Response timeline

  • Acknowledgement: within 48 hours
  • Initial assessment: within 5 business days
  • Fix timeline: severity-dependent; critical issues are prioritised immediately
  • Disclosure: coordinated with the reporter

Security researchers are credited in release notes unless anonymity is preferred.


High-Priority Areas

Given SENTINEL's purpose, these components are especially critical:

Audit log integrity — Vulnerabilities enabling undetected modification or deletion of tamper-evident audit logs undermine legal defensibility. Treat as critical.

Federation privacy — Vulnerabilities causing the federation layer to transmit raw message content or user identities (rather than encrypted behavioral signatures only) are a serious privacy breach. Treat as critical.

Fairness gate bypass — Vulnerabilities allowing a model that failed demographic parity checks to deploy are high severity.

API authentication — Unauthorized event ingestion or risk score retrieval affects safety decisions made by platforms relying on SENTINEL.


Out of Scope

  • Vulnerabilities in third-party dependencies (report upstream)
  • Theoretical attacks without a practical exploit path
  • Social engineering
  • Issues requiring physical server access

Security Architecture Notes

  • Audit logs use cryptographic chaining — tampering with historical entries is detectable
  • The federation layer operates on encrypted behavioral signatures; raw message content never leaves the originating platform
  • All API endpoints require authentication; see docs/api/authentication.md
  • The fairness gate runs as a pre-deployment check; models failing parity tests are rejected before serving predictions

SENTINEL is not a substitute for professional security review of your platform. Platforms remain responsible for their own security posture.

There aren't any published security advisories