A published document describing what a model is, how it was built, what it was tested on, and where it is known to fall short. The AI equivalent of a datasheet or an ingredients list. Their quality varies considerably, and they are increasingly expected by regulators.
Technical view
Typically covers intended and out-of-scope uses, training data at a high level, evaluation results including disaggregated performance across groups, known limitations and safety testing performed. Commercial sensitivity means details of training data and methodology are often thin, which limits how far independent assessment can go. Related artefacts include datasheets for datasets and system cards covering deployed applications rather than models alone.
Related
Part of AI Terminology Index, section: Testing, Safeguarding and Running Models in Practice.