Only 15 of 10,000 AI Models Had Security Docs

A UK government review found security-relevant documentation on only 15 of the 10,000 most-downloaded Hugging Face models with permissive licences—fewer than 0.2%—and just 131 of roughly 2,000 AI-related GitHub repositories contained substantive security content, the study reports. It puts a number on what model cards leave unsaid: a team selecting an open-weight model has roughly a one-in-667 chance of finding security information in that sample, so provenance, vulnerability review, and deployment tests remain buyer work; the archive’s open-weight adoption analysis tracks why distribution does not equal production readiness.