A standardized disclosure document accompanying an AI model that describes its intended use, training data, evaluation results, limitations, and known failure modes.
Definition and scope
Model cards originated in Mitchell et al. (2019) 'Model Cards for Model Reporting' (FAccT). The pattern was adopted by Hugging Face Hub (default model template), Google PAIR, and Microsoft Responsible AI. EU AI Act Art. 53 codifies model-card-style disclosures for general-purpose AI models — providers must document training-data summary, capabilities, limitations, intended use, and evaluation methodology. NIST AI RMF (Govern 1.3, Map 5.1) cites model cards as a transparency mechanism. ISO/IEC 23894 (AI risk management) endorses analogous documentation. Distinguish from: (a) 'system card' — wraps a model card with deployment-context information (OpenAI uses this term for GPT-4 family); (b) 'data sheet' — Gebru et al. 2018, focuses on training datasets rather than models; (c) 'fact sheet' — IBM's term for similar disclosure. Model cards remain voluntary in most jurisdictions; the EU AIA Art. 53 disclosure is the first binding equivalent.
Used by these instruments
Related concepts
- Frontier-Tier AI— A categorical classification of AI models above certain capability or compute thresholds, indicating
- Systemic Risk (AI)— A regulatory designation indicating that a general-purpose AI model poses risks of significant scale
- Red-Team Evaluation— Structured adversarial probing of an AI model's capabilities and behaviour before deployment, design
Appears in topic articles
Editorial note
When comparing model cards across providers, normalize for completeness: cards may omit training-compute, dataset composition, or evaluation methodology under trade-secret claims. EU AIA Art. 53 carves out trade-secret exemptions narrowly.
References
Take this further — sign up free
Save, compare, or get alerts when Model Card changes. Policy Window is the analyst workbench layered on top of this wiki — free for researchers, civil society, and verified policymakers.