Provenance review is the process of checking where a model came from, how it was obtained, and how it is being used. In cybersecurity, provenance helps defenders verify whether a model is legitimate, sanctioned, and aligned with policy before it is trusted in production or on an endpoint.
This matters because model distribution changes the trust boundary. A model downloaded from an unofficial source may be tampered with, repackaged, or paired with unsafe tooling. In attacks, provenance gaps can let malicious or modified models enter private environments, where they may be used for phishing, code generation, or automated abuse. In defense, provenance review supports procurement checks, allowlists, signature verification, access logging, and environment hardening. It is especially important for open-weight models, where once the weights are copied, control shifts from the original developer to the deployer.