sensedefence/hex repository overview

⁠Sense Defence Hex – AI Model Security Scanner

Hex is an advanced AI model scanning tool developed by Sense Defence⁠. It performs static analysis of machine learning model files to detect:

• Known CVEs and vulnerable versions
• Embedded secrets (API keys, credentials)
• Unsafe layers (e.g., Lambda, custom operators)
• Malicious pickle or joblib code
• Serialization exploits
• Dangerous function usage (eval, exec, os.system, etc.)

⁠Supported Model Formats

• Keras (.h5)
• TensorFlow SavedModel
• PyTorch (.pt, .pth)
• Pickle / Joblib
• Scikit-learn
• XGBoost
• ONNX

⁠Use Cases

• Security validation before model deployment
• Supply chain risk prevention
• Secure CI/CD and MLOps pipelines
• AI model compliance & auditing

⁠Quick Start

docker run --rm -v $(pwd):/models sensedefence/hex --input my_model.onnx -o result.json

• --input: File name inside the mounted /models directory
• --output: Output JSON report (optional). Defaults to console output.

⁠Output Report

The tool generates a structured JSON report including:

• identified_framework – model type and version
• vulnerabilities – CVEs found
• secrets – API keys or tokens detected
• status – safe, caution, or unsafe
• verdict – Security verdict
• risk_score – Numerical severity rating

⁠Example

docker run --rm -v /home/user/models:/models sensedefence/hex --input model.onnx -o result.json

Output saved to /home/user/models/result.json

© 2025 Sense Defence AI. All rights reserved.