A model card is a reference document for an AI model. It describes what the model is intended to do, where it came from, how it was evaluated, known limits, and any licensing or usage constraints. In security terms, it is part of the model’s provenance: the evidence that helps a team decide whether the model is trustworthy enough to deploy.
Model cards matter because reused AI assets can hide risk. A model may look functional while still carrying unsafe assumptions, weak testing, or a license that blocks a commercial or operational use. Defenders use model cards to compare versions, verify intended use, and spot gaps before production. Attackers benefit when this documentation is missing, incomplete, or misleading, because teams are more likely to import poisoned, unsupported, or misconfigured models without proper review.