ModelScan
A pre-load scanner that flags suspicious constructs in supported machine learning model files.
Published by Protect AI
How to use it
Run the scanner as part of model intake before a third-party artifact is loaded by an inference service or opened in a developer environment. Review any findings and require approval for exceptions.
Keep scan results tied to the artifact digest. A clean result helps with triage and belongs alongside other controls such as provenance checks, isolated processing, and an allowlist of model formats.
- Access
- Open-source software
- License
- Apache-2.0