Model provenance
Evidence of where a model file came from and that it has not been altered since its publisher released it: publisher identity, hashes, signatures and a verifiable base model.
Provenance is to models what a software bill of materials is to code. It matters because fine-tunes, merges, quantised builds and adapters are routinely redistributed under the original name, and because backdoors can be implanted in a few weights. Practical controls are a private model registry, hash pinning, publisher signing (Microsoft Foundry and others have announced plans) and a weight diff against the declared base before deployment.
Related terms
Closed modelConfidential computingNVFP4Open-weight modelSecret scanning
Get the brief in your inbox
One email every morning: the stories that mattered in AI, each with a one-line "why it matters" for your role. Free, no ads, unsubscribe any time.
Published daily by AI Briefing. We never share your address.