Study: Current Model Cards Insufficient for Downstream Governance of Open-Weight Models

Model cards are documents that provide information about an AI model such as purpose, data, and limitations, used as transparency artifacts in model repositories.

The study found that most cards lack critical information such as model heritage, training data, and observed safety behaviors, making them less useful for developers to assess risks.

The framework combines three components: model cards, acceptable use policies, and licenses, to integrate informational, normative, and legal dimensions.

Organizations in the region should adopt stronger governance practices when deploying open-weight models, verifying cards, policies, and licenses before use.