EU AI Act glossary
Model Card
A model card is a short structured document that describes a machine learning model's intended use, performance characteristics, limitations, and evaluation results. It is a precursor to — but not a substitute for — EU AI Act Annex IV technical documentation.
Last updated 17 June 2026
What is a model card?
A model card is a concise, standardised document that communicates the key facts about a machine learning model: what it does, how it was trained and evaluated, where it performs well, and where it fails. The term was introduced by Google researchers (Mitchell et al., 2019) and has become a widely adopted practice in the ML community.
Model cards are commonly published alongside open-source models (Hugging Face, ModelHub) and are increasingly required by enterprise buyers during vendor evaluation.
Model card vs. Annex IV technical documentation
Model cards and Annex IV files serve related but distinct purposes:
| Dimension | Model card | Annex IV technical file |
|---|---|---|
| Audience | Developers, data scientists, ML community | Regulatory authorities, notified bodies, enterprise security reviewers |
| Format | Informal, usually markdown or PDF | Formal, structured per legal specification |
| Legal status | Voluntary (outside specific regulations) | Mandatory for high-risk AI under Article 11 |
| Scope | Typically a single model version | Entire AI system lifecycle including monitoring and post-market |
| Evidence | Summary statistics, qualitative descriptions | Evidence-linked artefacts traceable to your live systems |
| Updates | Ad hoc, when the team remembers | Required whenever a substantial modification occurs |
A model card is a good starting point. It is not a substitute for Annex IV documentation — an enterprise security reviewer or notified body will not accept it in place of a complete technical file.
What a model card covers that feeds into Annex IV
Many elements of a well-written model card directly correspond to Annex IV sections:
- Intended use → Annex IV §1 (general description, intended purpose)
- Training data → Annex IV §3 (data governance)
- Evaluation results → Annex IV §4 (validation and testing)
- Out-of-scope uses → Annex IV §1 (foreseeable misuse)
- Ethical considerations → Annex IV §2 (design specifications, assumptions)
modeldocs ingests your existing model card content and maps it to the correct Annex IV section, then fills the gaps with artefacts pulled from your live ML systems.
→ See Technical Documentation for what evidence-linking means, or run the Readiness Check to see which gaps remain.
Check your Annex IV coverage
Nine questions. Two minutes. See exactly which sections of your technical file are missing evidence.
Run the free readiness check