European Union - AI Training Data Transparency (Regulation (EU) 2024/1689)

Overview

The European Commission adopted the "Explanatory Notice and Template for the Public Summary of Training Content for general‑purpose AI models" on 24 July 2025 to implement the transparency obligations contained in Article 53(1)(d) of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689). The Template provides a mandatory, standardised format for public summaries that GPAI model providers must publish when placing a GPAI model on the Union market or putting it into service in the Union. It sets out three core information blocks: model/provider metadata; an organised listing of the main data source categories used in training (public datasets, licensed datasets, crawled/scraped online content, user data, synthetic data, and other sources); and processing/governance aspects addressing copyright, removal of illegal content and compliance with data protection obligations. The Template is intended as a "minimal baseline" to improve comparability, help rights‑holders exercise remedies, and enable market surveillance while providing limited allowances for trade secrets where justified. For the official explanatory page and downloads see the Commission’s AI Office library entry (Explanatory Notice and Template — Shaping Europe’s digital future).

Definitions

The Template uses terms aligned with the AI Act and accompanying guidance. "General‑Purpose AI (GPAI) model" denotes models meeting the Act’s definition for systems intended to perform a wide range of tasks and that can be adapted for downstream uses. "Training content" / "training data" cover the input data, content and artifacts used at any stage of model development, including pre‑training and subsequent post‑market training iterations. The Template distinguishes categories of data sources: "publicly available datasets", "private/licensed datasets", "data obtained through crawling/scraping", "user data", "synthetic data" and "other data". "Provider" means the natural or legal person that develops and places the GPAI model on the market or puts it into service in the Union, including entities modifying an upstream model where they become the provider under the Act. "Material change" is any change in data composition or training regime that could reasonably affect the information contained in the published summary.

Governance and Institutional Framework

The Template sits within the AI Act implementation architecture: the European Artificial Intelligence Office (AI Office) within DG CONNECT hosts the GPAI portfolio and provides the secretariat for Union‑level coordination; the European Artificial Intelligence Board (Board) supports consistent application across Member States and facilitates cooperation between national competent authorities and market surveillance authorities. National competent authorities designated under the AI Act will be responsible for local enforcement and market surveillance. The Commission’s publication was accompanied by guidance documents and a Q&A to support consistent use of the Template; see the Commission downloads and supporting materials available from the AI Office library and press material (Commission AI Office — Template page). The Board and the AI Office will coordinate responses to "qualified alerts" and advise on referrals. Where necessary, the Commission has powers to request documentation, technical access for evaluations and to impose corrective measures and fines under Article 101 of the AI Act.

Key Focus Areas

The Template focuses on transparency across several interlocking areas. First, provenance and scope: providers must disclose the main categories of content used and provide indicative size ranges for modalities (for example, token ranges for text or approximate numbers of images), enabling stakeholders to understand scale and modalities without requiring full datasets. Second, source details: for crawled data providers must include narrative descriptions and report the top domains or domain groups contributing to the dataset (to the extent feasible and not prejudicial to security or legitimate confidentiality). Third, legal and rights governance: the Template requests confirmation of licensing arrangements for private datasets, measures for respecting text and data mining opt‑outs, and procedures used to identify and remove illegal content. Fourth, privacy and data protection: the Template asks providers to summarise how personal data has been handled and, where appropriate, whether Data Protection Impact Assessments or other safeguards were applied. Fifth, update and lifecycle management: the Template requires periodic updates and specifies conditions that trigger immediate publication updates (material changes). These focus areas aim to help rights holders, researchers and authorities identify potential legal or ethical issues while limiting technical details that could reveal trade secrets or compromise model security.

Implementation Framework

The Template is the legally mandated format that GPAI providers must use to prepare the public summary required by Article 53(1)(d) of the AI Act. The Commission set the effective dates such that the Template applied to GPAI models placed on the market on or after 2 August 2025; models already on the market before that date have a transitional period to comply by 2 August 2027. Providers must publish the summary in an accessible manner, include model identifiers and contact points for rights‑holders and regulators, and maintain a version history and date of last update. The Template recommends the frequency and level of updates (at least every six months for active models and upon material changes). Where disclosure could affect public safety, ongoing investigations, or legitimate trade secrets, providers may redact limited elements but must provide a justification and a non‑redacted description of the omitted information to the competent authority on request. Complementary instruments—such as the voluntary GPAI Code of Practice and the Commission’s Guidelines for providers—offer implementation guidance and examples of compliant summaries (See Commission downloads and guidelines).

Monitoring and Evaluation

Monitoring will be carried out by national competent authorities, the AI Office and market surveillance authorities, coordinated through the European Artificial Intelligence Board. The AI Act grants authorities powers to require documentation, request access for technical evaluations and verify the accuracy of published summaries. The Commission may act on complaints and on "qualified alerts" referred by the Scientific Panel of independent experts or other competent bodies. The Template is intentionally concise to allow automated or manual comparability checks; however, it is designed as an initial transparency tool rather than a full audit record — authorities retain powers to request the technical documentation and records that underlie the public summary for enforcement and investigative purposes.

Penalties, Liability, and Appeals

Non‑compliance with the Template and the AI Act’s GPAI obligations can trigger enforcement action. Under the AI Act providers of GPAI models face specific administrative fines (Article 101): in certain cases fines of up to EUR 15 million or up to 3% of annual worldwide turnover (whichever is higher) may be imposed. Other enforcement measures include corrective orders, temporary suspension or withdrawal from the market, and publication of decisions. Providers have procedural rights (notification of preliminary findings and the right to be heard) and may challenge Commission decisions before the Court of Justice of the European Union. The Template itself requires providers to maintain records and cooperate with authorities to speed up investigations and reduce liability risks.

Relationship to Other Instruments

The Template complements and implements obligations set out in the AI Act and interacts with other EU legal frameworks. It should be read alongside the AI Act’s technical documentation requirements (Annex XI/XII) and transparency obligations, the Commission’s Guidelines for GPAI providers, the voluntary GPAI Code of Practice and relevant sectoral rules (for example, copyright law, the General Data Protection Regulation and consumer protection rules). Where personal data is concerned, the template disclosures must be coherent with data protection obligations and, where relevant, providers should reference Data Protection Impact Assessments and cooperation with data protection authorities. The Template is also designed to interface with national market surveillance procedures and the EU database for high‑risk AI systems when relevant.

International Alignment

The Commission framed the Template in the context of international engagement on AI transparency and interoperability. The Template’s approach — standardised public summaries with graduated levels of detail — aims to enable cross‑jurisdictional comparability while respecting differing legal regimes on trade secrets and national security. The Commission’s AI Office has been active in technical dialogues (for example with US and other partners) to align approaches to provenance, watermarking and content authentication. The Template therefore contributes to international policy coherence while remaining firmly anchored in Union law and enforcement mechanisms (AI Office international dialogue).

Implementation Timeline

Sources and References