System Card
A safety evaluation document released alongside an AI model that systematically records the model's capability boundaries, potential risks, red-team testing results, and abuse mitigation measures. System cards are considered an important component of industry transparency practices.
Core Facts
Timeline (last 90 days)
Model Cards and System Cards, promoted by companies such as OpenAI and Google, require that models be released with accompanying documentation of known limitations and risks.
System cards are technical documents that AI companies attach when releasing new models, describing the model's capability boundaries, limitations, safety evaluation results, and potential risks.
All Facts (3)
The System Card is a document that OpenAI has attached to model releases since GPT-4, originating from the "Model Card" concept proposed by Google researchers in 2018.
65%UnverifiedModel Cards and System Cards, promoted by companies such as OpenAI and Google, require that models be released with accompanying documentation of known limitations and risks.
50%UnverifiedSystem cards are technical documents that AI companies attach when releasing new models, describing the model's capability boundaries, limitations, safety evaluation results, and potential risks.
50%