Enterprise platform for safe LLM evaluation and defensive cybersecurity governance.
Built to support the safe development of frontier AI models (Mythos++ class), inspired by Anthropic’s Claude Mythos Preview System Card.
- Defensive Cyber Evaluation Engine with multiple specialized verifiers
- Strict Safety Gates — automatic rejection of offensive or harmful content
- Composite Reward Scoring combining accuracy, safety, calibration, and patching quality
- Full Audit Trail — all evaluations stored with detailed results
- Secure Sandboxing using gVisor
- Scalable Async Processing via Celery + Redis
- Production-ready deployment with Traefik + Let's Encrypt SSL
VulnerabilityScannerVerifierCyberAntiHackingVerifierOverEngineeringDetectorPatchVerifier
git clone https://github.com/Kubenew/mythos_safe_enterprise.git
cd mythos_safe_enterprise
cp .env.example .env
# Edit .env with your settings
docker compose -f docker-compose.yml -f docker-compose.override.yml up -d
# Run migrations
docker compose exec api alembic upgrade head
# Test the system
./test_curl.sh
Access:
API Docs: http://localhost:8000/docs
Celery Flower: http://localhost:5555
🏗️ Production Deployment
Bashdocker compose -f docker-compose.prod.yml up -d
See PRODUCTION_DEPLOYMENT_CHECKLIST.md for detailed steps.
📁 Project Structure
backend/app/verifiers/cyber_defensive/ — Core safety logic
backend/app/services/verification_service.py — Main evaluation service
backend/app/worker/ — Celery tasks
docker-compose.*.yml — Multiple deployment profiles
test_cases/ — Sample vulnerable code for testing
🛡️ Safety Philosophy
All evaluations are strictly defensive. Any attempt to generate exploits, payloads, or harmful content results in immediate rejection (composite_reward = 0.0).
This platform aims to help develop AI systems that strengthen defenders while minimizing dual-use risks.
📄 Documentation
GETTING_STARTED.md
DEPLOYMENT.md
ARCHITECTURE.md
PRODUCTION_DEPLOYMENT_CHECKLIST.md
Status: Production-ready scaffold with strong defensive cyber capabilities.
Built with responsibility, transparency, and safety at the core.