The AI Act and IP rights: key obligations for providers of general-purpose AI models

Belgium

The European Union’s Regulation 2024/1689, also known as the AI Act, is one of the first laws governing the deployment of artificial intelligence tools. This legislation categorizes AI systems into different risk levels and imposes corresponding obligations. In this last article in our series on AI and intellectual property rights, we discuss the provisions of the AI Act that will have an impact on your intellectual property.

1.    Copyright policies regarding general-purpose AI models

According to the AI Act, providers of general-purpose AI (“GPAI”) models must implement policies demonstrating their compliance with EU laws on copyright and related rights (Art. 53.1(a)).

A general-purpose AI model is defined as an AI system that is trained with a large amount of data using self-supervision and that is capable of competently performing a wide range of distinct tasks. AI models used solely for research, development or prototyping before they are placed on the market are excluded from the definition in the AI Act.

The providers of such general-purpose AI models are the entities that develop an AI system and put it on the market under their own name or trademark.

The copyright law policy of general-purpose AI providers should demonstrate how they will comply with the reservation of rights expressed by rights holders. As explained in one of our earlier Law-Now articles, the European Copyright in the Digital Single Market Directive (Directive 2019/790) has created an exception to the exclusive rights of the copyright owners for “text and data mining”, which is defined as “any automated analytical technique aimed at analysing text and data in digital form in order to generate information which includes but is not limited to patterns, trends and correlations”. Under this exception, general-purpose AI providers may reproduce or extract protected content for the purposes of text and data mining, but only if the rights holders have not expressly reserved the use of the protected content “in an appropriate manner, such as machine-readable means in the case of content made publicly available online”.

The AI Act does not make any changes to these provisions, and paragraph 105 of its recitals confirms that the text and data mining exception is applicable to GPAI providers, but those providers must now clearly document how they will comply with the obligation to respect the reservation expressed by the copyright owners. This obligation is applicable, regardless of the jurisdiction in which the copyright-relevant acts take place.

2.    Information policies for AI systems implementing general-purpose AI models

Art. 53.1(b) of the AI Act stipulates that the providers of general-purpose AI models must provide up-to-date information policies for customers that will integrate the general-purpose AI models into their AI systems. The policies must enable providers of AI systems to have a good understanding of what the general-purpose AI model can or cannot do. The following items must be included:

  • details about the tasks that the model is intended to perform and the types of AI systems into which it can be integrated;
  • an outline of the acceptable use policies that apply to the model;
  • information on how the model interacts with hardware or software that is not part of the model itself;
  • the architecture of the model and the number of parameters it contains;
  • the versions of relevant software related to the use of the general-purpose AI model, where applicable;
  • the modality (such as text or image) and the format of the inputs and outputs;
  • the licence under which the model is released; and
  • the date of release and the methods of distribution for the model.

The GPAI provider can draft this policy taking into account its own IP rights, confidential information or business secrets.

3.    Training data policies regarding general-purpose AI models

GPAI providers must also set out a policy detailing the content used for (pre-)training their models (Art. 53.1(d)), allowing rights holders to know whether their data might have been used in the training of these models. The recitals to the AI Act underline that this policy should provide a clear overview, and should not be too technically detailed. We propose implementing the following elements:

  • data collections used when training the relevant model;
  • which organisation has granted you permission to use this data, and when this was granted; and
  • censors where relevant as to not divulge trade secrets or confidential information.

The AI Office (the EU body responsible for the implementation and enforcement of the AI Act) will most probably enforce the obligations regarding training data policies and the copyright policies (see nr. 1 above) without checking every individual element of training data (Preamble, N° 108). The Office will also publish a template for training data policies, allowing for an efficient drafting of this overview.

While most parts of the AI Act will become applicable on 2 August 2026, all the provisions referenced above will take effect on 2 August 2025.

4.    Conclusion

The AI Act marks a significant step towards comprehensive AI regulation, ensuring that AI development and deployment adhere to stringent standards. By mandating transparency regarding copyright compliance, training data and more, the EU legislator aims to protect intellectual property rights without disproportionately burdening the entities active in the AI sector. As the AI landscape continues to evolve, the AI Act will play a crucial role in shaping responsible and ethical AI practices within the EU.

This article was written by Arthur Sabbe, with the valued assistance of AI. This is the last article in the series on “AI and intellectual property rights”. All the articles are available on our website. If you have any further questions or comments, please do not hesitate to reach out to us at [email protected].