On 24 July 2025, the European Commission published a template for the public summary of the training content used by general-purpose AI models, in accordance with Article 53(1)(d) of the AI Act, which has been in force since 2 August 2025.
EU - European Commission - Explanatory notice and template for the public summary of training content for general-purpose AI models
Anno 2025
All providers of general-purpose AI models, including open-source models, are required to publish a sufficiently detailed summary of the data used to train their models, as provided for under Article 53(1)(d) and Recital 107 of the AI Act.
This summary pursues several objectives:
- Increasing transparency regarding the data used to train models, including copyright-protected content;
- Ensuring the effective application of European data protection rules, while also giving relevance to consumer rights and their protection;
- Identifying possible biases and adopting corrective measures to prevent discrimination;
- Promoting the dissemination of information and encouraging scientific research;
- Ensuring more competitive and transparent markets.
The summary must cover all stages of training and include all types of data used. The Commission provides a template intended to serve as a common basis, consisting of three sections:
- General information about the model;
- A list of the sources from which the data used were derived;
- The most relevant aspects of data processing.
The template must be completed in a comprehensive, narrative, and sufficiently precise manner, although technical details are not required; providers may choose to provide additional information on a voluntary basis.
The template seeks to balance transparency with the protection of providers’ trade secrets, in compliance with the Charter of Fundamental Rights of the European Union. The data requested are therefore limited to what is necessary to ensure the effective exercise of the rights protected under EU law.
In particular, different levels of transparency are required depending on the source of the data.
For licensed data, only minimal disclosure is required. For private data obtained from third parties, a list is required only if the data are already publicly known; otherwise, they may be described in general terms. Public datasets, on the other hand, require a greater level of detail. For data obtained through scraping from online sources, disclosure of the most relevant information is required, such as the tools used, the collection period, online sources, and so on. Minimal information is required with regard to data obtained from user interactions and synthetic data. In any event, an exact indication of the composition of the data sources is not required; only a general estimate of their size is needed.
The template must be completed in a simple and intelligible manner. The information must be provided in good faith and in a comprehensible way, and a degree of flexibility is always guaranteed so that only relevant information needs to be indicated.
The AI Office may verify providers’ compliance with the guidance, impose corrective measures, and issue penalties. The use of mediation tools or Alternative Dispute Resolution (ADR) mechanisms is encouraged for the resolution of any disputes.
Where a party modifies existing general-purpose AI models, that party in turn becomes a provider and is required to publish information only on the data used for the modification.
The summary must be updated every six months, or earlier in the event of substantial changes. A single summary may be used for different models if they are identical, while in the case of different versions, only the data that differ from the original model need to be indicated.
Providers must publish the summary of the data used at the time the model is placed on the market, on their own website and through all distribution channels. The disclosure obligation has been in force since 2 August 2025, while models already on the market must comply by 2 August 2027. In the event of difficulties in retrieving the data, providers must provide an adequate justification.
The AI Office’s monitoring activities will begin on 2 August 2026.
The Commission reserves the right to update the template provided, depending on future needs.
The text of the document is available at the following link and in the download box.
Elena Frassinelli, UNITN
Pubblicato il: Sabato, 24 Maggio 2025 - Ultima modifica: Giovedì, 14 Maggio 2026
