System card
A detailed report a lab publishes with a major model, describing its safety testing, risks and safeguards.
Also: safety card
System cards grew out of model cards as frontier labs began releasing longer reports alongside major launches. OpenAI's GPT-4 system card in 2023 and later cards from OpenAI, Anthropic and Google describe red-teaming, tests of dangerous capabilities in areas such as biology and cybersecurity, evaluations by outside groups and the safeguards applied. Critics have noted that some cards arrived weeks after launch or not at all, and several governments have asked labs to commit to publishing them.