model-supply-chain-security

To secure AI model artifacts against supply chain attacks, implement artifact signing, provenance attestation, and vulnerability scanning. This ensures model weights and dependencies are not maliciously altered before deployment.

Provenance attestation for AI model weights verifies the origin and build process of every model artifact. It provides proof that the build pipeline was not compromised and validates artifact integrity.

Detect poisoned pretrained weights in your build pipeline by enforcing dependency controls and vulnerability management workflows. Scanning runtime images and attesting to artifact provenance mitigates the risk of compromised tools.

Yes, this approach generates an SBOM for AI model dependencies. SBOM workflows are implemented alongside artifact signing to detect and mitigate risks from compromised dependencies and vulnerable runtime environments.

Yes, you can use these controls to implement trusted model promotion in CI/CD. The pipeline automatically signs model artifacts, attests to their provenance, and scans for vulnerabilities before promoting them to production.

Artifact signing for runtime image security verifies integrity but does not automatically patch vulnerable dependencies. It must be combined with continuous vulnerability management and dependency controls to address insecure runtime images.