Blog post by Stephen Wyber
Last year, we published a blog that highlighted concerns about an emerging trend in the contracts offered by publishers to libraries, whereby publishers seek to close down possibilities to carry out text and data mining (TDM). This is despite the explicit permission for this provided by Article 3 of the Directive on Copyright in the Digital Single Market, at least as concerns non-commercial research.
Unfortunately, there is no sign that this trend is slowing down in the context of a debate that, for all its intensity, seems to pay little attention to the impact of copyright and related practices on research.
This blog updates and complements our piece from last year, drawing on new examples. The title is, admittedly, dramatic, but points to the risk that a set of behaviours from the rightholder sector could undermine the EU’s ability to deliver on a specifically European approach to AI based on content subject to copyright.
Just Say No: A Blanket Approach to the Use of AI Tools
As already highlighted in our previous blog, publishers have started to attempt to block all uses of AI tools by researchers, even the ones that come as standard on any computer.
A recent blog from the University of Colorado gives an idea of a situation that many libraries are likely facing at the moment – demands from a publisher that exclude the use of their materials in Word or Excel with Copilot, or even from making posters on Adobe Pro or Canva, given that they have embedded AI features.
The University of Colorado blog sets out the library’s efforts to come to a reasonable solution, which were met with refusals from the publisher. The library ultimately decided it could not sign a contract with terms that were simply unenforceable.
Of course, it is impossible to determine what the motivation of the publisher might have been, but their approach is doubly harmful–either it means that researchers are prevented from using even the most basic AI tools to support productivity, or there is no access at all. Neither supports the EU’s ambitious goals in science and research.
Foreclosing Markets: Barriers to Entry for Analytical Tools
A second issue that is emerging concerns the efforts by some publishers to control uses of works for TDM/AI Training in order to ensure that they alone can develop downstream AI tools.
This risk has been highlighted in the case of Beck in Germany, which has a very strong position in the market for legal scholarship. This blog underlines concerns that this behaviour leads to a situation where Beck alone is able to develop AI tools based on legal content. Its large market share means that competitor publishers would not be able to match it, let alone external players looking to develop products.
The situation here raises parallels with the concerns that the Digital Markets Act (DMA) sought to address–the possibility for major platforms to enjoy an unfair advantage in downstream markets because of the data that they hold. It certainly stands in the way of the development of a strong European start-up ecosystem in this area.
Undermining Open Access: Rolling Back CC-by Licences
Next is a trend highlighted in the Technical University of Delft blog, which calls out a drive by some publishers to push researchers to shift from CC-BY to CC-BY-NC licences, on the pretence of this being about ‘protecting’ works.
Of course, what such a move protects is not the researcher, but rather the publisher’s ability to licence works to AI training companies. While some indeed have sought to control their data in order to protect the possibility of developing additional tools, others have been happy to sign agreements with Generative AI companies.
Such a model works well for those players able to raise enough funding on markets in order to pay the fees publishers request, but risks creating a barrier to entry for all other stakeholders.
Meanwhile, it is not only commercial AI training that CC-BY-NC licences prevent, but also any other use in the context of most research-industry partnerships, as well as by start-ups or scale-ups. This not only runs counter to widely accepted definitions of open access, but also to the EU’s own desire to facilitate broader access to knowledge.
Say It Isn’t So: Fake It Til You Break It
A final tactic appears to be based on simply implying that the law is what you want it to be. This is most obvious in the efforts by organisations acting in the name of authors in Sweden to demand that they receive a cut of the money allocated to support the development of a Swedish large language model based on the collections of the National Library there.
While such claims have a weak legal basis, they are at risk of working due to the disproportionate political pressure that rightholders can apply, as well as an aversion to risk and controversy among public institutions.
Similarly, there are repeated efforts to claim that the TDM exception in the Digital Single Market Directive does not cover AI training. We have addressed this before, and the recent Gema vs OpenAI court decision confirms that the training of large language models is covered by the exception, as does this analysis by Open Futures. Infringement only occurs if a model produces output that is substantially similar to a copyrighted work—a distinction that is hardly new.
Proponents of this argument have likely been encouraged by the proposals made by Axel Voss in his draft report on copyright and generative AI. Voss not only repeats claims about the limitations of the existing TDM exception, but also implies that those training AI should not wait for a change in the law, and rather just start to pay immediately. While this report carries no legal force, such a stance raises concerns regarding adherence to the rule of law.
Moreover, its status as a European Parliament report could mislead people into assuming it represents a formal interpretation of the law, which it does not. Given the draft’s failure clearly to exempt research from its scope, this confusion–and the behaviours it may justify–undermines efforts to pursue the EU’s goals around AI.
A Common Thread?
Drawing a line between these different behaviours by various publishers and rightholders would be speculative at best, and would likely over-estimate the degree to which the publishing industry is unified in its response to AI.
Nonetheless, there are certainly commonalities in the impact that these behaviours have. Together, they serve both to limit the possibilities for research and other institutions to contribute to the development of European AI, as well as to the ability of European researchers to use AI to advance science.
In this, they run directly contrary to the EU’s ambitions, both as concerns AI in science and science for AI. Critically, they also represent an approach to copyright that fails to take account of the basic principle of balance – none of the uses being prevented above represent an unjustified prejudice to legitimate interests.
It is time to recognise this clearly, and simply say no to practices that undermine Europe’s interests.
4 December 2025
