Model Card
Standardized documentation providing key information about an AI model.
How the laws define this
Regulators and advisory bodies consistently define a model card as a standardized document or artifact summarizing an AI model's intended purpose, training data characteristics or provenance, performance metrics, and known limitations to ensure transparency, enable auditing, and inform procurement. While every source highlights these core descriptive elements, the instruments diverge on specific required disclosures and legal force. Instruments like the Draft AI Act presented by political party 'Da, Bulgaria' and the Draft Principles for AI Legislation explicitly require model cards as formal, mandatory technical documentation for high-risk systems, incorporating elements such as model architecture, versioning, or specific risk mitigations. Conversely, policy frameworks such as the Artificial Intelligence Playbook for the UK Government, the NITI Aayog Approach Document, the MCIT Principles and Guidelines, and the UAE AI Office AI Adoption Guideline treat model cards as recommended guidance or standard artifacts to support accountability, with some adding distinct details like maintenance needs or fairness evaluations.
Synthesised from the 8 statutory definitions below. Each one is quoted in full, with its source.
How jurisdictions differ
What is distinctive in each definition, binding law in force first. Showing 6 of the 8 instruments that define this term; all are listed under Definitions below.
| Jurisdiction | Instrument | Status | What is distinctive |
|---|---|---|---|
| United Kingdom | Artificial Intelligence Playbook for the UK Government | In Force (Amended) | Adds operational controls alongside risk mitigations and emphasizes procurement assurance and auditability for UK government use. |
| Thailand | NSTDA AI Ethics Guideline | In Force | Specifies model cards to support transparency, informed deployment, and monitoring. |
| United Arab Emirates | AI Adoption Guideline in Government Services (UAE AI Office) | In Force | Includes known risks alongside limitations and focuses on procurement and auditing for UAE government services. |
| Bulgaria | Draft AI Act presented by political party 'Da, Bulgaria' | Draft | Mandates maintenance and provision on request, includes versioning, and ties to conformity assessments under proposed Bulgarian law. |
| Ecuador | AI and Data Protection Regulation Bill | Under Review | Adds fairness considerations and recommended deployment constraints under Ecuador's proposed bill. |
| Thailand | Draft Principles for AI Legislation | Draft | Specifies use for high-risk systems, includes architecture and mitigation measures, under Thailand's draft principles. |
What it triggers
- In the UK Government's AI Playbook, model cards are recommended to document key model information including controls.source
- Under Thailand's NSTDA AI Ethics Guideline, model cards are recommended to support transparency, informed deployment and monitoring.source
- In the UAE AI Adoption Guideline in Government Services, model cards are used to support transparency, procurement, and auditing.source
- Under Bulgaria's Draft AI Act, providers must maintain and provide model cards on request to support transparency, procurement checks, and conformity assessments.source
- Under Ecuador's AI and Data Protection Regulation Bill, model cards would describe key model details to support transparency, explainability, and accountable use.source
Definitions (8)
A required piece of model documentation that describes an AI model's purpose, intended use, training data characteristics, performance metrics, limitations, risk mitigations, and versioning; providers must maintain and provide model cards on request to support transparency, procurement checks, and conformity assessments.
A standardized document that describes an AI model's intended purpose, training data characteristics, performance metrics, known limitations, fairness considerations, and recommended deployment constraints to support transparency, explainability, and accountable use.
A standardized form of documentation that summarizes an AI model's purpose, training data, performance metrics, limitations, and appropriate use cases to improve transparency and enable meaningful explanations for users and auditors. The Approach Document endorses model cards as part of documentation standards for responsible procurement, deployment, and audit readiness.
A concise, standardized document describing a model's intended use, performance across relevant metrics, known limitations, evaluation results, and provenance information. Model cards are required by the Guideline to support transparency, informed deployment and monitoring.
A standardized form of technical documentation that summarizes a model's provenance, architecture, training data characteristics, evaluation metrics, intended uses, known limitations, and mitigation measures; required for transparency, audits, and conformity assessments of high‑risk systems.
A standard artefact recommended by the Playbook that documents a model's intended use, training data provenance, performance metrics, known limitations, risk mitigations and operational controls to support transparency, procurement assurance and auditability.
A standardized artifact describing an AI model's intended use-cases, training data provenance, performance metrics, known limitations, fairness evaluations, and maintenance needs to support transparency, procurement, and auditing.
A standardized documentation artifact that summarizes a model's intended purpose, characteristics of training and evaluation data, performance metrics, limitations, and known risks to support transparency, procurement, and auditing of AI models.
Not to be confused with
Model cards are standardized summary documents with specific contents like intended use and limitations, while technical documentation covers broader compliance records.
Model cards provide structured documentation on model details to enable transparency, whereas algorithmic transparency refers to disclosing decision-making processes.
Questions and answers
What must a model card include under the UK AI Playbook?
It documents intended use, training data provenance, performance metrics, known limitations, risk mitigations, and operational controls.
When are model cards required under draft laws?
Bulgaria's Draft AI Act would require providers to maintain and provide them on request; Ecuador's bill would standardize them for transparency.
Related Terms
General-Purpose AI (GPAI)
AI models trained on broad data capable of performing diverse tasks across domains....
Technical Documentation
Detailed records demonstrating AI system compliance with regulatory requirements....
Algorithmic Transparency
The disclosure of how AI systems make decisions and process information....