Your understanding of the supply chain risk is correct for the container layer, but it's far more granular. A Docker image from a "trusted" publisher can still pull in a base layer with a vulnerable libc, or a build script that fetches dependencies from an unverified mirror at build-time. The image is just the final artifact.

You're right to ask for practical tools. Image scanners like Trivy or Grype are a start, but they only catalog known CVEs. They won't catch a malicious commit in a downstream GitHub dependency that hasn't been assigned a CVE yet. For your local AI models, the risk extends to the training data, the framework binaries, and even the Python packages used to run the inference.

A beginner checklist is a good idea. It should include steps for generating a Software Bill of Materials (SBOM) for your containers, pinning hashes for all dependencies, and implementing build-time policies that fail on unsigned artifacts. The real challenge is managing that process without it becoming a full-time job.

Yeah, that's basically it. The random Docker image or model is the obvious one. I get nervous just thinking about all the layers in those things.

Your point about beginner checklists is exactly what I'm hoping for too. I get lost once people start talking about SBOMs and attestations. A simple "start here" list for someone with a Pi and a few containers would be a lifesaver.

Do you think they'll cover ways to check those random GitHub repos, beyond just looking at star counts? That's my biggest blind spot.