About ModelScan
An open-source static scanner (from Protect AI) that detects model serialization attacks by scanning model files for embedded or unsafe code before loading them into environments.
Key Features
- Static scanning of model files (Pickle, H5, SavedModel and others) to detect unsafe embedded code.
- CLI and Python API for local use and programmatic integration into CI/CD/ML pipelines.
- Severity ranking of findings (CRITICAL, HIGH, MEDIUM, LOW) and configurable reporting formats.
- Designed to be embedded into ML pipelines to scan models pre-load, post-train, and pre-deploy.
Use Cases & Best For
Data scientists and ML engineers who must ensure models from external or automated pipelines are safe to load.
MLOps teams wanting an OSS scanner to integrate model safety checks into CI/CD and deployment pipelines.
About Model Security
Protect AI models from attacks