Model cards were introduced as a voluntary habit: a page describing training data, intended use and known limitations. New rules in the EU and parts of Asia reference them directly, and lawyers are now reviewing wording that engineers used to write in an afternoon.
What changes in practice
Claims about capabilities must be supportable, limitations must be specific, and evaluation methods must be reproducible on request. Several labs have quietly rewritten older cards. Expect the documents to get longer and more cautious.
Key takeaways
- Regulators now cite model cards in guidance
- Capability claims need evidence on file
- Older cards are being revised
This story is part of our ongoing coverage of ai. We will update it as more details emerge.
