Copyright uses for AI training go beyond the scope of the text and data mining exception: Several EU member states agree

This acknowledgment reinforces the arguments of the authors' coalition, which points to a problematic application of the Directive that allows AI models to be trained on copyrighted content without the permission of the creators.

By Ilias Maroutsis
January 21, 2025

Unclarities regarding the applicability of the Text and Data Mining (TDM) exception to AI training admit many Member States of the European Union as shown by their responses to a questionnaire on the relationship between generative artificial intelligence and copyright.

The questionnaire was circulated to the 27 Member States last summer by the Hungarian Presidency of the Council of Europe, recognising that the increasing prevalence of AI-generated content poses significant challenges to the creative industries. The Member States were invited to submit proposals on seven different topics related to AI and copyright.

The insights from the recently published Member States' replies seem to support the claims of the Coalition of Creators, who sent an open letter to the relevant EU Commissioners in December, pointing to the problematic reality regarding AI and authors' rights that has resulted from a misinterpretation of the 2019 Directive on the TDM exception.

The signatories of the open letter claimed that all generative AI models in existence today are trained in complete opacity on vast amounts of copyrighted content and personal data, scraped and copied from the internet without permission or compensation to the creators. This is because the relevant EU directive of 2019 introduced an exception to copyright for text and data mining (TDM), unless creators and other rights holders have expressly reserved their rights (Article 4 of the CDSM Directive). According to the coalition of creators although Article 4 was adopted years before the sudden emergence of generative AI technologies and does not mention or define "artificial intelligence" and "generative AI", the European Commission appears to have retrospectively interpreted this exception broadly to cover the systematic and extensive use of creators' protected works and performances without any authorisation. 

The responses of the Member States

Based on the responses to the questionnaire, it is clear that the relationship between AI and copyright is at the centre of attention and that several activities are taking place in the Member States to analyse the links between AI and copyright.

While most of the Member States' contributions showed that the current EU legal framework is sufficient to address the issues that arise, the majority of Member States raised practical issues where, in their view, more clarity and certainty would be needed to better implement the existing EU acquis.

According to the report, the most frequently raised issue in this context was the applicability of the text and data mining (TDM) exception and its opt-out mechanism, introduced by Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019 on copyright and related rights in the digital single market (hereinafter: the DSM Directive), to the AI training process. According to a significant number of contributions, there are certain uncertainties regarding the applicability of the TDM exception to AI training. This was reflected in a number of national positions and many stakeholder views. In this context, several Member States considered that copyright uses for AI training go beyond the scope of the TDM exception.
However, Member States agreed that there is no need to provide new or additional copyright-related protection. Several Member States expressed the view that the existing legal framework, including the current opt-out provision of Article 4(3) of the DSM Directive, combined with the AI Act's transparency obligation with respect to training data, should be sufficient to encourage the conclusion of licences between rightholders and AI companies. In their view, however, it is crucial that these provisions are properly implemented and enforced.

A significant number of Member States and stakeholders considered that the mass scale of AI-related copyright uses and the practical challenges for rightholders in policing them both underline the importance of collective management in this area. Many Member States reflected the views of their stakeholders, who supported the individual licensing approach and licence-based education for generative AI.

With regard to further measures that could be taken at EU level to facilitate the conclusion of licences between right holders and AI developers, there were diverging views on the possible establishment of a remuneration scheme for generative AI activities.

You can read the whole report here

--

Photo credit: mikemacmarketing
Photo source