Short answer
Providers of general-purpose AI models must publish a summary of what they trained on, using a template issued by the AI Office:
"draw up and make publicly available a sufficiently detailed summary about the content used for training of the general-purpose AI model, according to a template provided by the AI Office." —
Three things define the obligation:
- It is public. Not a filing to a regulator — a document made publicly available.
- It follows a mandatory template. The AI Office's template is the operative specification. Its current version should be taken from the Commission's own page rather than from any secondary description.
- It is a summary, not a dataset disclosure. The Regulation's recitals are explicit that it should be comprehensive in scope rather than technically detailed.
The obligation applies to all GPAI providers, including those established outside the EU. Article 53(2) exempts certain free and open-source models from points (a) and (b) of Article 53(1) — but not from point (d).
What "sufficiently detailed" means
Recital 107 supplies the interpretive standard and is worth reading closely, because it frames both the purpose and the limits:
"In order to increase transparency on the data that is used in the pre-training and training of general-purpose AI models, including text and data protected by copyright law, it is adequate that providers of such models draw up and make publicly available a sufficiently detailed summary of the content used for training the general-purpose AI model. While taking into due account the need to protect trade secrets and confidential business information, this summary should be generally comprehensive in its scope instead of technically detailed to facilitate parties with legitimate interests, including copyright holders, to exercise and enforce their rights under Union law, for example by listing the main data collections or sets that went into training the model, such as large private or public databases or data archives, and by providing a narrative explanation about other data sources used." —
Four operative elements emerge.
Purpose: enabling rights enforcement. The summary exists so that rightholders and others with legitimate interests can work out whether their material may have been used and act on it. That purpose is the test of adequacy — a summary so vague that no rightholder could form a view is unlikely to satisfy it.
Comprehensive in scope, not technically detailed. Breadth over depth. All significant categories of training data should be covered, without exhaustive technical specification.
List the main data collections or sets. The recital names "large private or public databases or data archives" as the kind of thing to be listed.
Narrative explanation for other sources. Where sources cannot be listed individually, a narrative account is contemplated. Recital 107 also says the template should be "simple, effective, and allow the provider to provide the required summary in narrative form".
Trade secrets are protected — but do not swallow the duty. The recital requires trade secrets and confidential business information to be taken into due account, while still requiring a generally comprehensive summary. Confidentiality shapes the level of detail; it is not a basis for withholding the summary.
What it is not
Not a work-by-work disclosure. Recital 108 draws the line explicitly:
"the AI Office should monitor whether the provider has fulfilled those obligations without verifying or proceeding to a work-by-work assessment of the training data in terms of copyright compliance. This Regulation does not affect the enforcement of copyright rules as provided for under Union law." —
Not a licence or a copyright defence. Publishing the summary does not make the training lawful. Recital 108's final sentence preserves copyright enforcement entirely. A provider can be fully compliant with Article 53(1)(d) and still face infringement claims.
Not the same as the technical documentation. Article 53(1)(a) requires technical documentation containing at least the Annex XI information, provided to the AI Office and national competent authorities on request. Article 53(1)(b) requires information and documentation for downstream providers integrating the model, containing at least the Annex XII elements. Those are targeted disclosures; the training-content summary is public.
Not a substitute for the copyright policy. Article 53(1)(c) separately requires a policy to comply with Union copyright law and to identify and comply with Article 4(3) reservations, including through state-of-the-art technologies. Publishing a summary does not discharge it.
Who must publish, and who is exempt
All GPAI providers placing a model on the Union market, irrespective of establishment, under Article 2(1)(a). Recital 106 confirms that the copyright-related obligations apply regardless of where the training acts took place.
Open-source models are not exempt from this obligation. Article 53(2) exempts models released under a free and open-source licence, with publicly available parameters, architecture and usage information, from points (a) and (b) only — and that exemption does not apply at all to models with systemic risk.
Proportionality for smaller providers. Recital 109 states that compliance should be commensurate and proportionate to the type of provider, excludes the need for compliance for persons developing or using models for non-professional or scientific research purposes, and provides that compliance should take due account of the provider's size and allow simplified ways of compliance for SMEs, including start-ups, without excessive cost.
Third-country providers must additionally appoint an EU authorised representative under Article 54, who must verify that the Article 53 obligations have been fulfilled and keep the Annex XI documentation available for ten years.
Codes of practice and standards
Article 53(4) allows providers to rely on codes of practice within the meaning of Article 56 to demonstrate compliance until a harmonised standard is published. Compliance with European harmonised standards gives a presumption of conformity to the extent those standards cover the obligations.
A provider that adheres to neither must "demonstrate alternative adequate means of compliance for assessment by the Commission".
Article 53(7) provides that information and documentation obtained under Article 53, including trade secrets, is treated in accordance with the confidentiality obligations in Article 78.
Practical implications
Capture provenance at ingestion. The summary cannot be reconstructed from a trained model. Source categories, dataset identities, acquisition routes and licence positions must be recorded as data is collected.
Design the summary around its purpose. If a rightholder reading it could not begin to assess whether their material was used, it is probably not sufficiently detailed.
Keep it current. Models are retrained and fine-tuned. The summary should reflect the model as placed on the market, and be updated as that changes.
Coordinate with the copyright policy. The summary and the Article 53(1)(c) policy draw on the same underlying records. Build one provenance system serving both, plus the Annex XI documentation.
Use the official template. It is the operative specification, and it is subject to revision by the AI Office. Take the current version and any accompanying FAQ directly from the Commission's page rather than relying on descriptions of it.
Watch the timetable. The GPAI obligations have their own application dates and the Regulation has been amended since adoption. Confirm which obligations apply to your model and from when.
Operational checklist
- Confirm whether your model is a general-purpose AI model placed on the Union market.
- Record data provenance at ingestion, by category and by named dataset.
- Obtain the current AI Office template and complete it in full.
- Cover all significant data categories; prefer breadth over technical depth.
- List main data collections and give a narrative account of other sources.
- Apply trade-secret protection to the level of detail, not to the existence of the summary.
- Publish it, and keep it accessible and current.
- Keep the Annex XI technical documentation separately, available on request.
- Maintain the Article 53(1)(c) copyright policy alongside it.
- Appoint an EU authorised representative if established in a third country.
Conclusion
Article 53(1)(d) requires a public, template-based summary that is comprehensive in scope rather than technically detailed, written so that rightholders can exercise their rights. It is not a dataset dump, and recital 108 confirms the AI Office will not conduct work-by-work copyright assessment.
The practical difficulty is upstream of the document. A provider that did not record where its data came from cannot produce an adequate summary afterwards. Build the provenance record at ingestion, and the summary, the copyright policy and the technical documentation all become outputs of the same system.
Sources
This article provides general legal information, not advice for a specific matter. Laws, facts and procedural deadlines may change the result. Follow the linked primary sources and obtain local advice before acting.