Blog post by Stephen Wyber
Amongst the many reports about AI being released at the moment, the European Union Intellectual Property Office (EUIPO)’s most recent one on Generative AI and copyright is nonetheless likely to get more attention than most.
This blog – part of a series on recent developments around AI – explores the report and what it does (and doesn’t) cover.
Understanding the EUIPO’s Perspective on Generative AI and Copyright
It takes as a starting point the possibility to create a market for licensing AI training that is established by Article 4 of the Copyright in the Digital Single Market (CDSM) Directive. This provision allows rightholders to opt-out of having their material used for Text and Data Mining (and so AI training), even when a beneficiary of the exception has lawful access. This stands in contrast to Article 3, which allows for training and analysis on any work an organisation has legal access to, but only on a narrow non-commercial basis.
Having set this out, the bulk of the report then works within the parameters of this existing legislation, exploring how to operationalise the opt-out, and so guarantee the foundations of such a market. What it does not do is question the effectiveness of licensing as a way of achieving policy goals, rather simply assuming that the key enabler of a successful AI sector in Europe is getting the market right.
A key giveaway may be in the title of the report – The Development of Generative AI from the Perspective of Copyright. As highlighted in a previous blog, this can tend to lead to an evaluation of the status quo, thus ignoring the issue of how best to achieve wider policy goals. As a result, the focus is either on how to monetise copyrighted materials optimally.
The report therefore offers a valuable overview of how a licensing market might be built in order best to serve the interests of rightsholders. It offers useful insights into how to ensure that those with smaller catalogues are not left out, and also recognises different preferences between creative sectors. Helpfully, it clarifies that (despite the arguments made by some) the use of materials to train AI is definitely covered by the DSM Directive.
Much of this is likely to have been down to the mandate given when the report was commissioned. What this means, however, is that it does not offer a serious consideration of what framework will best support research, development and innovation in Europe as it is done today.
Considering the Implications for European Research and Collaboration
First of all, there is the assumption that the additional costs created by a licensing scheme (over and above what has already been paid to access works in the first place) will not fundamentally place EU researchers at a disadvantage compared to those in other countries. It is surely wishful thinking to assume that this will not drive researchers and research-intensive players abroad.
Secondly, the report repeats (without providing any sort of evidence) the story sometimes shared by proponents of licensing that AI companies are using partnerships with research institutions as a loophole to avoid the application of Article 4, and so licensing fees.
This flies in the face of the EU’s long-standing focus on supporting and promoting research-industry collaborations (through the knowledge valorisation agenda). It also implies that research institutions themselves are not applying due internal process, including research ethics.
As such, it seriously undervalues the extensive and high-quality work being done in Europe’s research institutions to build and apply AI, and seemingly threatens any such institution engaging in partnerships with being accused of being a ‘front’ for nefarious purposes. It also raises another barrier to the EU’s stated goal of promoting research-industry collaborations, already made nearly impossible by the very narrow drafting of Article 3.
Towards Smarter AI Policy: Moving Beyond a Zero-Sum Game
Ultimately, the report therefore needs to be seen clearly as providing a deep but narrow perspective on questions around copyright and AI.
It is becoming increasingly urgent to disconnect the debate about how to support the cultural industries in the face of competing AI outputs from discussions about research. To avoid a zero (or even negative) sum game between promoting AI research and uptake and the promotion of the cultural industries, we need a less blunt approach.
KR21’s Principles on Artificial Intelligence, Science and Research offer a way forwards here, which we will be exploring in more depth in coming blogs.
Read other recent publications about copyright, research and AI:
- Focusing on the Future: Why AI Needs Freedom to Flourish in the UK
- A False Dichotomy, a Real Risk – Review of the AI Continent Proposals
- Boosting AI innovation in the UK
- Undermining Scientific Research
- Exploring the Intersection of AI, Text and Data Mining, and Copyright: A Reflection on the Stockholm Conference
13 May 2025
