The EU AI Act and training data

How the EU AI Act's general-purpose AI rules reach training data, including data licensed from US companies.

Updated
In short
  • Obligations for providers of general-purpose AI models applied from August 2, 2025.
  • Providers must publish a summary of training content using a template the European Commission published in July 2025.
  • Providers must have a copyright policy, including respecting text and data mining opt-outs.
  • The AI Office's enforcement powers, including fines, apply from August 2, 2026.

Who it applies to

The AI Act's rules on general-purpose AI (GPAI) models apply to providers placing those models on the EU market, wherever they are based. A US lab whose model is offered in the EU is covered. A US company that licenses data to that lab is not a provider, but the provider's obligations flow back into the license.

Key dates

DateWhat happens
August 2, 2025GPAI obligations apply
July 2025Commission publishes the training content summary template
August 2, 2026AI Office enforcement powers, including fines, apply

Source: European Commission.

The training content summary

Providers must publish a sufficiently detailed summary of the content used to train the model, following the Commission's template, including data sources (European Commission). Like California's AB 2013, this means buyers need a disclosable description of licensed data.

Providers must put in place a policy to comply with EU copyright law, including identifying and respecting rights reservations under the text and data mining exception in Article 4 of the Digital Single Market Directive (EUR-Lex). Rights holders can reserve their works from text and data mining in a machine-readable way. Licensed data sidesteps that question because the license itself grants permission.

What it means for US data sellers

  • Expect EU-facing buyers to ask for a description of your data they can include in their summary.
  • Expect warranties that you hold the rights you grant, and that the data does not include content you know was reserved from text and data mining.
  • If the data includes personal information about people in the EU, the GDPR applies separately.

What it means for AI teams

Keep a data manifest and license for every dataset. The same records support the EU summary, AB 2013 and buyer due diligence.

Questions

Does the EU AI Act ban training on copyrighted data?

No. It requires GPAI providers to have a copyright policy and respect text and data mining opt-outs, and to publish a training content summary. Licensing is one way to show permission.

When can the EU fine GPAI providers?

The European Commission says the AI Office's enforcement powers, including fines, apply from August 2, 2026.

Sources

  1. European Commission: AI Act regulatory framework
  2. EUR-Lex: Directive (EU) 2019/790 on copyright in the Digital Single Market

Tell us what you need

Book a free 30-minute consultation. Tell us what data you want to buy or license, and we'll tell you honestly whether we can help and what it involves.

  • We reply to every enquiry within two working days.
  • No upfront fees for sellers
  • We never hold your raw data