Opinion by the European Copyright Society on Generative AI

The European Copyright Society (ECS) has written an Opinion on Copyright and Generative AI. The ECS argues that the balancing exercise needed in the high stakes AI environment should be guided by the following principles:

• The interests of human authors and performers;
• The interests of users and of the wider public, anchored in the fundamental rights framework established by the Treaties and the Charter of the European Union;
• The enhancement of research and innovation.

The Opinion identifies tensions relating to model training, innovation policy and author remuneration, including rights reservation, transparency requirements and jurisdiction. These issues are playing out globally. While the legal analysis is located within the framework of EU law, UK policy makers too will have to reflect on copyright reform in this wider context.

Specifically, the ECS considers that the current development of Generative AI, under the regulatory framework set up by the Directive on Copyright in the Digital Single Market (CDSM) of 2019 and the AI Act of 2024 (Regulation (EU) 2024/1689), requires responses to six top line issues:

1. The determination of the scope of the text and data mining (TDM) exception: the exception enacted in Arts. 3 and 4 of the CDSM Directive at a time when the Generative AI development could not have been fully anticipated, can be interpreted as covering some operations of training of a Generative AI model, but certainly not all aspects or stages of the life cycle of AI models and systems, from curating a dataset for training to the generation of an image, text or other media, by users. The exact scope of the TDM exception, and hence the copyright status of acts carried out at each stage of development and operation of Generative AI models and systems, should be further studied and analysed. That would require a decision as to whether acts of reproduction or public communication occur and which actors are liable for such acts. Under such assessment, the possibility of commercial use of models trained for scientific research and the effect of the exercise of the opt-out provided by Art. 4 CDSM Directive, on the availability of lawfully accessible sources for the research exception provided by Art. 3 CDSM Directive, merit particular attention.
2. The content of the obligation under Art. 53(1)(c) of the AI Act related to the reservations of rights: in particular, the technologies that can be used to express the opt-out should be identified and regularly reviewed; the rightholders entitled to opt-out and the opt-out modalities, including the timing and the location, should be clarified.
3. The scope and modalities of the transparency obligation laid down by Art. 53(1)(d) of the AI Act: in particular, the relevant information to be included in the summary and the impact of the transparency obligation on the assessment of the lawful access criterion contained in Arts. 3 and 4 CDSM Directive should be clarified.
4. The privileges for Research and for Open Source models: the importance of research and the key role of open source data and software in the field of AI should guide the interpretation of the CDSM Directive and the AI Act, and lead to needed clarification of some of their provisions, with the objective of preserving the fundamental rights of research, academic freedom and education. The uncertainties raised by the Hamburg court decision in the LAION case, as to the interface between Art. 3 and Art. 4 of the CDSM Directive, should particularly be addressed in order to avoid general purpose AI (GPAI) model providers relying on training for the purposes of research, hereby escaping the more restrictive frame of the exception of Art. 4.
5. The articulation between the CDSM Directive and the AI Act: the CDSM directive is a private law instrument organizing a protection of private rights on a territorial basis, whereas the AI Act is a public law that regulates the safety of AI products, as a condition for importation and use in the EU. That raises several issues in the articulation of both legislative texts, notably the territorial scope of the obligations imposed, the entities covered by the different obligations, the effect of the AI office’s voluntary Code of Practice, the distinct modes of enforcement of the obligation laid down by the CDSM Directive and by the AI Act. These points should be clarified.
6. The fair remuneration of authors and performers for all acts of exploitation of their works and performances occurring in the life cycle of Generative AI models and systems (including when an opt-out from the application of Art. 4 CDSM Directive has been exercised and when their works or performances are included in a dataset that has been licensed to an AI provider) needs to be reaffirmed as a fundamental principle of the EU acquis. The Commission should look at the best ways to ensure such a remuneration, including remuneration rights or other compensation mechanisms, in concert with Member States.

The full Opinion on Copyright and Generative AI is available as a CREATe working paper and on the ECS website.

Members of the ECS drafting committee were Prof. Séverine Dusollier (Sciences Po Paris), Prof. Martin Kretschmer (CREATe, University of Glasgow), Prof. Thomas Margoni (Centre for IT & IP Law (CiTiP), KU Leuven – and a member of CREATe), Prof. Peter Mezei (University of Szeged),  Assoc. Prof. João Pedro Quintais (IViR, University of Amsterdam) and Prof. Ole-Andreas Rognstad (University of Oslo).

The European Copyright Society (ECS) is currently chaired by CREATe’s director Prof. Martin Kretschmer and will hold its annual conference on 16 May 2025 at the University of Glasgow. If you are tempted to explore beautiful Scotland in May, our Spring School (organised in cooperation with the reCreating Europe consortium) coincides with the ECS conference and is open for registration under the theme “(re)Humanising Creativity”.