Is AI Training Covered by the TDM Exception? A Legal Debate Affecting the Text+ Community
Zur deutschen Fassung dieses Beitrags
Researchers working with text data are likely familiar with the relatively new exceptions for Text and Data Mining (TDM) introduced by the 2019 EU Directive on Copyright in the Digital Single Market. These exceptions were implemented in German law in mid-2021, with one key provision (§60d UrhG) enabling research institutions to conduct TDM on any content to which they have lawful access. While this change has not eliminated all legal uncertainties, it has significantly simplified many aspects of research involving text data.
The Text+ Working Group on Legal and Ethical Issues has dedicated considerable attention to this topic, producing two publications focused primarily on the implications of §60d UrhG for language research. However, a major legal question has recently emerged within the Text+ community: Does training language models require permission from rightholders, or is it covered under the TDM exception? In other words, is AI training considered TDM?
Conflicting Perspectives
In September 2024, a report by Dornis and Stober, commissioned by Initiative Urheberrecht, argued in no less than 217 pages that generative AI training does not fall under the TDM exception. The report garnered significant attention and was even presented before the European Parliament.
However, only weeks later, the Regional Court of Hamburg (LG Hamburg) ruled in the opposite direction, asserting that the reproduction of works for AI training purposes can indeed be covered by the TDM exceptions (see Kluwer Copyright Blog).
With these contradictory interpretations, it is no surprise that researchers in the Text+ community have felt uncertain about the legal status of AI training under the TDM exception.
The Text+ Working Group’s Position
To address these concerns, the Text+ Working Group on Legal and Ethical Issues has published an official statement on the matter. Based on careful legal and technical analysis, the group firmly concludes that AI training falls within the scope of the TDM exception, and therefore in principle does not require permission from rightholders.
The full statement is available on our website, and we encourage its use as a reference whenever this issue arises.
The Working Group welcomes all comments, differing opinions, and additional insights. If you would like to contribute to this ongoing discussion, please comment below or reach out to us at office@text-plus.org.
Image Credit: CC-BY-4.0 Rens Dimmendaal / Johann Siemens
OpenEdition schlägt Ihnen vor, diesen Beitrag wie folgt zu zitieren:
Paweł Kamocki (26. Februar 2025). Is AI Training Covered by the TDM Exception? A Legal Debate Affecting the Text+ Community. Text+ Blog. Abgerufen am 30. April 2025 von https://doi.org/10.58079/13dsc