EXPLAINER

Model Cards and Datasheets (AI Documentation)

standardfreshness: stablesources not yet verified

Model cards and datasheets are structured documents describing what a model or dataset is, how it was built and evaluated, intended uses and limitations — enabling informed and accountable use.

Short answer: A model card records intended use, training data summary, evaluation (including by subgroup), limitations, ethical considerations and update history. Datasheets do the same for datasets. They are widely used on model hubs and are increasingly expected for governance.

What is it?

A model card is like a nutrition label for an AI model: what it is for, what went into it, how well it performs, where it fails and what to watch out for. Datasheets do the same for datasets.

Also known as: model card, datasheets for datasets, system card, AI documentation, transparency note, model documentation, dataset card, model card template.

Technical definition

Mitchell et al. (2018) propose model cards with sections for model details, intended use, factors, metrics, evaluation data, training data, quantitative analyses, ethical considerations and caveats; Gebru et al. (2018) propose datasheets for datasets. Variants include system cards (covering deployed systems and safeguards) and transparency notes. Regulatory regimes such as the EU AI Act impose documentation duties that card-style documents can help satisfy but do not replace (EU AI Act).

Entity type: standard · Freshness class: stable.

When should I use it?

Whenever you release or adopt a model.

Implementation patterns

  • Generate a card from your evaluation pipeline.
  • Include subgroup metrics and known failure modes.
  • Version the card with the model.
  • Review cards before adopting third-party models.

Technologies and frameworks

Commonly associated: Hugging Face model cards, Model Card Toolkit, dataset cards. Names are examples of what exists, not recommendations, and projects change quickly.

Practical guide

  1. Use a standard template.
  2. Fill from real evaluations.
  3. List limitations honestly.
  4. Version and publish.

Common failure modes

  • Marketing text instead of evidence.
  • Out-of-date cards.
  • Missing limitations.
  • No data-provenance information.

Troubleshooting

  • Cannot tell if a model suits your use — Look for intended use, evaluation conditions and limitation sections; test yourself.

Contrasted with

  • AI Governance — Governance is the whole control system; model cards are one documentation artefact within it.

Questions this page answers

  • What is a model card?
  • What should a model card include?
  • What is a datasheet for a dataset?
  • What is a system card?

Typical goals: document a model for users and auditors; read a model card critically.

Sources and verification

Status: not independently verified. This entry was authored on 2026-10-06 from general technical knowledge, without live checking of the sources below. Treat every source as a lead to confirm (identifier, title, authorship, current version) before publication or citation.

Freshness class: stable — Core ideas that change slowly (maths, classic algorithms). Re-review yearly.

Canonical sources (unverified leads):

  • Mitchell et al., Model Cards for Model Reporting (2018) — arXiv:1810.03993
  • Gebru et al., Datasheets for Datasets (2018) — arXiv:1803.09010
Learning path: Governing AI (step 5 of 7)