Data cards in law

Who requires a data card? The two laws on our board

Updated

2 of the 6 standards and templates on the board are laws and 3 are voluntary formats; the texts read for Data Cards do not say which it is.

Talk to a specialist

The text the card answers to

California AB 2013 and the EU AI Act Article 53(1)(d) training-content summary

  1. Section: sources

    AB 2013 item (1) asks for the sources or owners of the datasets

  2. Section: personal information

    Open: find out before publishing, since AB 2013 and the Data Cards Playbook both ask

  3. Section: EU training-content summary

    Follow the AI Office template, which the Commission calls a common minimal baseline for the public summary

  4. Format of the card

    A README.md in the dataset repo, rendered as the card, with a YAML metadata section

Get this checked by a specialist

A buyer will ask for the card before the licence. Tell us about the dataset and we reply with what the texts on the board ask of it.

Has a buyer already made the data-holding company an offer for its data?

Does the data-holding company already license or sell this data?

Next: where to reach you. Free, no obligation.

How this calculator works

Which standards on the board are law

The 2 laws on our board are California AB 2013, which asks a developer of a generative AI system for a high-level summary of its datasets, and the EU AI Act. Its Article 53(1)(d) tells providers of general-purpose AI models to "draw up and make publicly available a sufficiently detailed summary about the content used for training" of the model, "according to a template provided by the AI Office". Of the other standards on the board, 3 are voluntary formats.

Source: European Commission AI Act Service Desk: Article 53, read for the data card generator

What the laws say about who must comply, and from when, is in their own text. The passages we rely on are quoted on each law's row of the board on our dataset page: AB 2013 from its bill text, and Article 53 as the Commission's AI Act Service Desk gives it. This page gives no view on whether either applies to you.

Ask about documenting a dataset you plan to license

A buyer will ask for the card before the licence. Tell us about the dataset and we reply with what the texts on the board ask of it.

Has a buyer already made the data-holding company an offer for its data?

Does the data-holding company already license or sell this data?

Tick the box above to send your enquiry.

Sources

This site's own dataset, every figure quoted from its source and dated:

  1. European Union (official text)
  2. Croissant Format Specification | Croissant site
  3. European Union (official text)
  4. [2204.01075] Data Cards: Purposeful and Transparent Dataset Documentation for Responsible
  5. The Data Cards Playbook - Data Cards Playbook
  6. NIST AI RMF
  7. Croissant - MLCommons
  8. Dataset Cards · Hugging Face

A buyer asked for a data card?

Pick the market and the format in the outline tool, and it shows the sections each text asks for.

Talk to a specialist