poojakira/Model-Supply-Chain-Auditor
GitHub: poojakira/Model-Supply-Chain-Auditor
Stars: 0 | Forks: 0
# Model-Supply-Chain-Auditor
[](https://github.com/poojakira/Model-Supply-Chain-Auditor/actions/workflows/ci.yml)
**ML Supply Chain Provenance & Security Auditor.** A specialized tool for verifying the integrity and safety of ML model artifacts (`.safetensors`, `.pt`, `.pkl`, `.ckpt`) before they enter your production environment.
## 🚀 Key Features
## 💻 CLI Examples
### 1. Basic Scanning
# Scan a SafeTensors model (Recommended)
python scan.py model.safetensors
# Scan a legacy PyTorch/Pickle model (Requires opt-in)
python scan.py model.pt --allow-pickle
### 2. Signing & Provenance
# Generate a keypair
python -c "from src.provenance.signer import generate_keys; generate_keys('my_model')"
# Sign a model artifact
python scan.py model.safetensors --sign my_model.pem
# Verify a model signature
python scan.py model.safetensors --verify my_model.pub
### 3. Policy-Based Audit
# Run scan against a specific security policy
python scan.py model.pt --policy policy.yaml
## 🔗 CI/CD Integration
Integrate Model-Supply-Chain-Auditor into your GitHub Actions to block insecure models at the gate:
jobs:
model-audit:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Audit ML Models
run: |
pip install -r requirements.txt
# Fail build if model is malicious or fails signature verification
python scan.py models/deploy.safetensors --verify keys/prod_pub.pem
- name: Upload Scan Results
if: always()
run: python scan.py models/deploy.safetensors --format sarif --output results.sarif
## 🛡️ Detection Coverage
| Threat | Strategy | Status |
| --- | --- | --- |
| **Pickle RCE** | Static Opcode Analysis | ✅ Detected |
| **Signature Fraud** | Ed25519 Verification | ✅ Prevented |
| **Metadata Injection** | Regex/Pattern Matching | ✅ Detected |
| **Archive Evasion** | Recursive ZIP extraction | ✅ Scanned |
## 📜 Documentation
- [SECURITY.md](./SECURITY.md) - Disclosure policy and security scope.
- [THREAT_MODEL.md](./THREAT_MODEL.md) - Assets, adversaries, and mitigations.
**Status**: Flagship Tool. Essential for SEC-624 compliance and secure ML supply chains.