I. STRUCTURE AND PURPOSE OF THE REGULATION
1) Regulation 2024/1689 established harmonised rules on AI. The purpose of the Regulation is to establish a uniform legal framework, in particular for the development, placing on the market, putting into operation and use of artificial intelligence systems (‘AI-systems’) within the Union, in a manner consistent with the values of the Union, to promote human-centric and trustworthy artificial intelligence (AI), while ensuring a high level of protection of health, safety and fundamental rights, as enshrined in the Charter of Fundamental Rights of the European Union (‘Charter’), including democracy, the rule of law and environmental protection, to provide protection against the harmful effects of AI-systems in the Union and to support innovation (recital
2) The Regulation entered into full application on 2.2.2026. However, the provisions concerning the so-called prohibited practices (Chapters I and II) have already been applied since 2.2.2025. While the provisions concerning General Purpose AI- GPAI (Chapter III Section 4, V, VII, XII, Article 78) since 2.8.2025. Finally, certain other provisions (Article 6, paragraph 1 and the corresponding obligations of the Regulation) since 2.8.2027. (Article 113).
II. INTELLECTUAL PROPERTY PROVISIONS
1) The Regulation generally sets the rules for the operation of IT systems. In this context, it states in article 53 that the providers of general-purpose AI models are required, among other things, to "implement a policy for compliance with the Union law on intellectual property and related rights, in particular for the identification and compliance, including through advanced technologies, with the preservation of rights expressed pursuant to Article 4(3) of Directive (EU) 2019/790".
2) Further recital 105 of the Regulation states that: "General purpose AI models, in particular large production AI models, capable of producing text, images and other content, present unique opportunities for innovation, but also challenges for artists, writers and other creators and the way their creative content is created, distributed, used and consumed. Developing and training such models requires access to vast amounts of text, images, videos and other data. Text and data mining techniques can be used extensively in this context to retrieve and analyze such content, which may be protected by copyright and related rights. Any use of content protected by intellectual property rights requires the permission of the relevant owner, unless relevant copyright exceptions and limitations apply. Directive (EU) 2019/790 established exceptions and limitations allowing reproductions and exports of protected works or other subject matter, for the purposes of text and data mining, under certain conditions. Under these rules, rightholders may choose to retain their rights over their works or other protected objects to prevent text and data mining, except for the purposes of scientific research. Where opt-out rights have been expressly reserved in an appropriate manner, providers of general purpose AI models need to obtain permission from rightholders if they wish to perform text and data mining from such works.
3) Continuing, recital 106 of the Regulation states: "Providers of general purpose AI models on the Union market should ensure compliance with the relevant obligations of this Regulation. To this end, providers of general purpose IT models should implement a policy of compliance with Union law on intellectual property rights and related rights, in particular to identify and comply with the preservation of the rights of beneficiaries as provided for in Article 4(3) of Directive (EU) 2019/790. Any provider that has a general purpose AI model on the Union market should comply with this obligation, regardless of the jurisdiction in which the intellectual property rights-related acts underlying the training of those general purpose AI models take place. This is necessary to ensure a level playing field between providers of general purpose AI models, where no provider should be able to gain a competitive advantage in the Union market by applying less stringent intellectual property rights standards than those provided for in the Union."
4) Furthermore, according to recital 107 of the same Regulation: "In order to increase transparency regarding the data used in the pre-training and training of general-purpose AI models, including texts and data protected by intellectual property law, it is appropriate that providers of such models compile and make public a sufficiently detailed summary of the content used to train the general-purpose AI model." Taking due account of the need to protect trade secrets and confidential business information, the summary should be generally comprehensive in its scope and not technically detailed in order to facilitate parties with legal interests, including intellectual property rights holders, to exercise and enforce their rights under Union law, for example by listing the main collections or datasets used to train the model, such as large private or public databases or data archives, and by providing a descriptive explanation of other data sources used. It is appropriate for the AI Service to provide a template for the summary, which should be simple and effective and allow the provider to provide the required summary in a descriptive format.
5) Recital 108 then states that: “As regards the obligations imposed on providers of general-purpose AI models to implement a policy of compliance with Union copyright law and to make public a summary of the content used for training, the AI Service should monitor whether the provider has fulfilled those obligations without verifying the training data or assessing it on a project-by-project basis for compliance with copyright. This Regulation is without prejudice to the enforcement of copyright rules as provided for in Union law.
6) Finally, recital 109 states that "compliance with the obligations applicable to providers of general-purpose AI models should be proportionate and proportionate to the type of model provider, excluding the need for compliance for persons developing or using models for non-professional purposes or for scientific research purposes, who should, however, be encouraged to comply voluntarily with such requirements. Without prejudice to Union copyright law, compliance with those obligations should take due account of the size of the provider and allow for simplified ways of compliance for SMEs, including start-ups, which should not entail excessive costs and should not discourage the use of such models. In the event of a modification or improvement of a model, the obligations of providers of general-purpose AI models should be limited to that modification or improvement, for example by supplementing the already existing technical dossier with information on the modifications, including new training data sources, as a means of complying with the value chain obligations set out in this Regulation.
III. LINKING THE REGULATION WITH OTHER EU INTELLECTUAL PROPERTY LEGISLATION
1) From the above references it follows that general-purpose TN models should not be developed at the expense of third-party intellectual property. This practically means that during their training, they should not use third-party intellectual property data without the express permission of these rights holders.
2) Text and data mining, which is extensively regulated in Directive 2019/790, is a technical process that can digitally retrieve huge volumes of texts, images, videos and other data. These data, often protected as third-party intellectual property, form the basis for pre-training and training general-purpose AI models.
3) Article 4(3) of Directive 2019/790, which is also explicitly invoked by the AI Regulation, clearly states that the exceptions allowing text and data mining are possible if “the use of works and other subject matter referred to in this paragraph has not been explicitly restricted by the rightholders in an appropriate manner, such as by machine-readable means in the case of content made publicly available online”.