08 / Trust & Governance
Model card
A short document describing a model: what it is for, what it was trained on, how it was tested, and where it should not be used. The label on the tin.
How you’ll hear it
“Is this model suitable for hiring decisions?”
“Read the model card. It lists that as an unsupported use.”
Avoid
Buying or deploying a model without reading its documentation.
The longer version
Model cards and similar documentation set out intended use, known limitations, evaluation results and risks. They are the starting point for procurement and risk assessment, and regulators increasingly expect this kind of documentation to exist.
In depth
A model card is a short, structured document that describes an AI model: what it was built for, what data it was trained on, how it was evaluated, how it performs across different groups, its known limitations, and uses it should not be put to.
It is the closest thing AI has to a product label. For anyone buying or deploying a model, it is the first document to read, because it tells you whether your intended use is one the builders tested and supported.
The quality of model cards varies a great deal. Some are detailed and honest about weaknesses; others say little. A missing or vague card is itself useful information.
Documentation of this kind is increasingly expected. The EU AI Act requires technical documentation for high-risk systems and information for those deploying general-purpose models.
What to do
- Ask for the model card or equivalent documentation before any procurement decision.
- Check that your intended use is listed as supported, and look closely at the limitations.
- Keep the documentation on file with your AI register.