Skip to content

Security: DmitrL-dev/AISecurity

Security

SECURITY.md

Security Policy

Reporting a Vulnerability

DO NOT open a public GitHub issue for security vulnerabilities.

Private Disclosure

Email: chg@live.ru

Telegram: @DmLabincev

Include:

  • Description of the vulnerability
  • Steps to reproduce
  • Affected component (sentinel-core, shield, brain, etc.)
  • Potential impact
  • Your suggested fix (optional)

Response Timeline

Stage Timeframe
Acknowledgment 48 hours
Initial assessment 1 week
Fix development 7-30 days
Public disclosure After fix is released

Process

  1. Acknowledgment — We confirm receipt
  2. Assessment — We evaluate severity and impact
  3. Communication — We keep you updated
  4. Fix — We develop and test a patch
  5. Release — We publish the fix
  6. Credit — We credit you (if desired) in commit history

Security Best Practices

API Keys

# Never hardcode credentials
import os
api_key = os.environ.get("SENTINEL_API_KEY")

Production Deployment

# 1. Set strong secrets in .env
cp .env.example .env
# Edit .env — change all placeholder values

# 2. Use docker-compose with restricted ports
docker-compose up -d

# 3. Run behind a reverse proxy (nginx/caddy) with TLS

Prompt Data

Sentinel scans prompts but does not store them by default. For privacy compliance, configure:

logging:
  include_prompts: false
  hash_only: true

Scope

This security policy covers:

Component Language Notes
sentinel-core Rust 61 detection engines — core attack surface
brain Python API backend — network-facing
shield C11 DMZ — memory safety critical
immune C EDR — kernel-level, highest severity
strike Python Red team tools — handle with care

Known Considerations

  • signatures/jailbreaks.json is ~51 MB. If you fork, consider Git LFS.
  • The strike/ directory contains real attack payloads (39K+). Treat as sensitive.
  • Docker images should not be exposed to public internet without a TLS reverse proxy.

Contact


Last updated: February 2026

There aren’t any published security advisories