Brussels wrote the form AI makers must use to list their data

The system

A rights holder who wants to know whether an AI music tool trained on their catalogue has, until now, had almost nowhere to look. The European Commission's AI Office published the Explanatory Notice and Template for the Public Summary of Training Content for General-Purpose AI Models on 24 July 2025, the document providers of general-purpose AI models must use under Article 53(1)(d) of the EU AI Act. The template is mandatory, not optional guidance: any provider placing a general-purpose model on the EU market, including open-source providers, must publish a summary in this format.

What the documents establish

The template requires three sections, as the accompanying questions-and-answers page describes them: general information on the provider, the model, and the types and rough proportions of training content by modality; a list of data sources by category, including public datasets, private datasets, scraped web data, user data and synthetic data, with more detail required for scraped sources, such as crawlers used, collection period, and the top 10% of domains scraped; and a section on data-processing steps relevant to rights holders, including removal of illegal content. The obligation applies from 2 August 2025 for new models, with AI Office enforcement possible from 2 August 2026, and providers of older models have until August 2027 to publish. Non-compliance can draw fines up to 3% of global turnover or €15 million, whichever is higher. The template followed a stakeholder consultation drawing over 430 responses in mid-2024, and it is designed to protect trade secrets while still letting a rights holder check whether text-and-data-mining conditions under the EU's Copyright Directive were respected.

Craft and rights

For a European composer, session musician or catalogue owner, this creates a first standard place to check whether a specific model discloses training on datasets that might include their work, and whether a crawler visited a site hosting it. It does not, on its own, create a right to object or a license fee; it is a disclosure obligation, and this report treats it as exactly that, an editorial distinction worth holding onto against any looser claim that the EU has now settled AI training consent. What it gives an artist is leverage: information to bring to a rights organization or a lawyer, not a settlement.

Outcomes and open questions

Because the obligation only fully applies from August 2025, and older models get until 2027, the practical test of this template is still ahead: how detailed the published summaries actually turn out to be, and whether the AI Office enforces gaps against non-compliant providers once its enforcement power begins in August 2026.

• Has the AI system you use published a training-content summary in this template's format yet?
• Does the summary name specific datasets or domains that plausibly include your catalogue?
• What would you do with a documented information gap the provider justified rather than filled?

A mandatory disclosure form is a floor, not a resolution, and its value to musicians depends on how rigorously providers fill it in and how closely anyone checks.