LinkedIn analytics tracking pixel for AIDOLS AI consulting website performance measurement
Governance & Risk

Model Card

A model card is a short, structured document that describes an AI model's intended use, performance, limitations, training data, evaluation results across demographic and operational subgroups, and known failure modes — the AI equivalent of a nutrition label.

Full definition

Introduced by Mitchell et al. (Google, 2019), model cards are now the de-facto disclosure standard. HuggingFace requires them, the EU AI Act mandates equivalent disclosures for high-risk systems, and Anthropic/OpenAI/Google publish them for major releases. A useful card includes: intended use, out-of-scope use, training data summary, evaluation metrics by subgroup, ethical considerations, and update history. AIDOLS includes a model card review in every governance engagement on /governance/.

Why it matters

Without model cards, downstream users cannot make informed deployment decisions and regulators cannot audit. With them, AI procurement evolves from PDF marketing decks to comparable, auditable disclosures.

Example

A health-tech company evaluates three vision models for diabetic-retinopathy screening; only one publishes subgroup performance by skin tone, exposing a 9-point sensitivity gap that disqualifies the other two.

Source & further reading

Primary source: Mitchell et al. — "Model Cards for Model Reporting" (FAT*) (2019).

Citation policy: this entry is part of the AIDOLS AI Implementation Glossary and may be quoted for research, journalism, and education with attribution to aidolsgroup.com/nl/glossary/model-card/.