AI Models

Generate SBOMs from AI and machine learning models to identify dependencies, frameworks, and potential vulnerabilities in your ML pipeline.

Supported Platforms

Why Scan AI Models?

AI models often contain:

• Framework dependencies - TensorFlow, PyTorch, JAX, etc.
• Data processing libraries - NumPy, Pandas, scikit-learn
• Serialization formats - Pickle files, SafeTensors, ONNX
• Embedded code - Custom layers, preprocessing functions
• Configuration files - Model cards, tokenizer configs

Scanning these components helps identify:

• Known vulnerabilities in dependencies
• Outdated or deprecated libraries
• Security issues in serialization formats
• Supply chain risks in model components

Scanning Workflow

1. Connect to model repository - Link your Hugging Face account or model URL
2. Select models - Choose which models to scan
3. Analyze dependencies - Safeguard identifies all components
4. Generate SBOM - Create a comprehensive software bill of materials
5. Continuous monitoring - Track new vulnerabilities over time

Best Practices

• Scan before deployment - Check models before pushing to production
• Regular rescans - Dependencies may have new vulnerabilities discovered
• Review model cards - Understand the model's training data and limitations
• Use SafeTensors - Prefer SafeTensors over Pickle for security

Next Steps

• Hugging Face Integration