A model card is a short standardized document that ships alongside an AI model to explain what it is, what it's good at, where it fails, and how it should — and shouldn't — be used. The concept was introduced in a 2019 Google paper and has since become common practice.

A typical model card covers intended use cases, known limitations and biases, training data details, and benchmark results across demographic groups or task types — the way a nutrition label helps you decide whether to eat something.

Model cards matter because a model that performs well on average can still fail badly for specific groups, languages, or domains. Without documentation, a developer might deploy a medical Q&A model on a population it was never tested on. As AI moves into hiring, healthcare, and finance, model cards are becoming a baseline expectation from regulators — and in the EU, something closer to a legal requirement for high-risk systems.