EU Introduces Mandatory Template for the Public Summary for AI models

Starting from 2 August 2025, a new requirement comes into effect in the European Union for providers of general-purpose artificial intelligence (GPAI) models: the publication of a publicly accessible summary of the data used to train the AI product.

Who is affected?

The new rule applies to all developers of GPAI models — both commercial and open-source. If such a product is made available on the EU market, a summary must be published:

  • No later than the date the model is placed on the EU market — for all new models.
  • By 2 August 2027 at the latest — for existing models already on the market as of 2 August 2025.

What must the summary include?

The structure of the summary is built around three key components:

  1. General Information
    • This section should include details about the developer and the model, as well as general information about the training materials — for instance, what types of data were used (text, video, audio), the approximate volume of such data and its main characteristics.
  2. List of Data Sources
    • This part should identify the sources from which the training data were obtained: publicly available datasets, restricted-access datasets, data collected from websites, user-generated data, or synthetically generated data.
  3. Data Processing
    This section addresses how the data are processed with respect to the rights of affected parties, in accordance with EU law. In particular, it must specify:
    • how copyright compliance is ensured,
    • how unlawful content is removed,
    • what measures are taken to safeguard legitimate interests.

Developers are also required to disclose whether information collected through user interactions across their services and products — including interactions with AI models — was used during training (without disclosing the personal data itself).

The obligation takes into account the need to balance transparency with the protection of confidential information. Developers are not required to disclose trade secrets where such disclosure would be detrimental to their business interests.

The summary must be updated at least once every six months and whenever the model is further trained in a way that affects its training dataset.

If a third party modifies the AI model, the modified version must be accompanied by a new summary, covering only the data used in the course of modification. The name of the modified model must be indicated, along with a reference to the original version.

Non-compliance and Liability

The summary must be published on the provider’s website and through all distribution channels of the AI model.

Failure to publish the summary by 2 August 2026 (for new models) may result in an administrative fine of up to 3% of the undertaking’s total worldwide annual turnover or €15 million — whichever is higher.

To support compliance, the European Commission has released a Template for the Public Summary of Training Content for general-purpose AI models.

Source: Official EU website

Author: Yepikhava Liudmila

Contact our lawyer for more details

Write to lawyer

Attention Journalists: Use of REVERA website materials in publications is only allowed with our written permission.