What does the EU AI Act require about training data, and how does it affect a data license?
Short answer. Article 53 of the EU AI Act requires providers of general-purpose AI models to publish a sufficiently detailed summary of the content used to train them, using a template the European Commission published in July 2025. The template asks for less detail on commercially licensed data than on public datasets, but providers say whether they licensed content. A company licensing records into such a model should agree how its data will be described and keep its own records of what was delivered.
General information, not legal advice.
The obligation in Article 53
The EU AI Act, Regulation (EU) 2024/1689, sets duties for providers of general-purpose AI models: large models that can do many different tasks and are offered in the EU. Article 53 lists them. Providers keep technical documentation, give information to companies that build on the model, put in place a policy to comply with EU copyright law, and publish a summary of the content used for training.
The copyright policy matters for licensing. EU law lets rights holders reserve their works from text and data mining, and a provider has to respect those reservations. Licensed material is the clean alternative, since the rights holder has agreed in writing.
The training content template
The European Commission's AI Office published the template and an explanatory notice on July 24, 2025. The Commission says use of the template is mandatory for the Article 53 summary. The summary has three parts: general information about the model and its training data, a list of data sources by category, and information on how the data was processed, including measures taken on copyright reservations and unlawful content.
The source categories include publicly available datasets, data crawled from the web, data from users of the provider, synthetic data, and commercially licensed or other private data obtained from third parties. Large public datasets are named one by one. For licensed and private data, commentators reading the template note that it asks for a narrative description and lighter detail, in recognition of commercial secrecy, together with a statement of whether content was licensed from rights holders.
Dates
The obligations for general-purpose AI models applied from August 2, 2025, for models placed on the EU market from that date. Models already on the market before then have until August 2, 2027. The AI Office gained powers to enforce the rules for these models from August 2, 2026. A provider that significantly changes an existing model, for example by further training, reports on the content used for that change.
What it means for a company licensing its records
If a company's records go into a general-purpose model offered in the EU, the provider will describe them in its public summary, most likely in the licensed data section and in general terms. The company can ask in the license how the data will be described and whether the company will be named, and can ask for notice before publication.
The template also asks how data was processed, including steps against unlawful content and copyright reservations. A seller helps the provider by delivering a clear manifest of what each file is, where it came from and what was removed. That manifest is the same document a lab asks for in due diligence, so it serves both purposes. A company in the EU has its own data protection duties as well, which our regions pages describe. Counsel in the EU and at home should confirm how both regimes apply to the specific set.
Follow-up questions
Does the summary rule apply to a model trained only on licensed company data?
If the model is a general-purpose model placed on the EU market, yes. The duty depends on the kind of model and where it is offered, not on the kind of data, so licensed data is described in its section of the template.
Is a US company that licenses data covered by the AI Act?
The Article 53 duties sit with the model provider. A US seller is not the provider, but its data may be summarized if the model is offered in the EU, and EU data protection law applies if its records hold information about people in the EU.
Can the license keep our company out of the public summary?
The license can set how the data is described and protect commercial terms, but it cannot stop a provider from meeting the law. Since the template allows general wording for licensed data, agreeing the description is usually workable.
Didn't find your question? Ask us directly.