Blog post by Stephen Wyber
Just before the summer, Knowledge Rights 21 held a webinar alongside colleagues from the Lisbon Council, Bruegel and the Center for Data Innovation. This focused on the manner in which the EU treats AI and copyright as a zero-sum game, and how a fresh approach would support creativity and enable research.
The speakers—David Osimo (Executive Director and Director of Research, Lisbon Council), Bertin Martens (Senior Fellow, Bruegel), Ayesha Bhatti (Head of Digital Policy, UK and EU, Center for Data Innovation) and Ben White (Bournemouth University and Knowledge Rights 21)—each shared perspectives grounded in their work and research.
Across their interventions, they urged a step back from knee-jerk reactions, based on narrow and simplistic perceptions of the issues on the table, the actors involved, and the tools available. Rather, they made the case for a more positive, confident approach.
Here are a few key takeaways from the discussion – watch the recording.
1) Artificial Intelligence Raises Questions That Traditional Copyright Struggles to Answer
By focusing on what can be learned from aggregated data, AI is at the heart of a structural shift away from traditional modes of working based on interactions with individual works. AI’s power as a research tool comes from its ability to go beyond what an individual person can do, looking across far wider collections of data and information as a whole. Moreover, AI-driven research analyses trends, themes, styles, facts and connections, none of which are protected by copyright law. These are scenarios that copyright was never created for, and as a result many issues arise.
2) Copyright Allows Rightsholders to Control Ideas and Knowledge That Is in the Public Domain
While of course copyright explicitly does not control facts, data, relationships and trends, by allowing for control of the “wrapper” through which they are shared, it creates significant monopolistic powers for rightsholders over information that otherwise resides in the public domain. As a result, Article 4 of the CDSM Directive or any commercial use in the UK of AI requires relicensing of works an organisation already has legal access to despite having legitimately acquired them in the first place – for example through buying a licence, because they are published open access, or because they constitute an openly accessible website. When it comes to the research sector (where royalties to authors hardly exist), the effect is to impose restrictions – or costs – on those trying to develop new ideas and knowledge, in order to award power and revenues to those holding existing knowledge.
3) The AI-Copyright Debate Ignores 96% of the Economy
When it comes to copyright, AI policy makers are focussing almost exclusively on the voices of intermediaries in the entertainment industry – a sector of the economy estimated to equate to 4% of the economy on average across Europe. However, this ignores the fact that AI is used across the whole economy. In jurisdictions where there is a wider perspective on what will benefit societies and the economy as a whole, more long-term and innovation-friendly policy developments are taking place around AI.
Indeed, this more friendly approach is not even necessarily at the cost of cultural industries. Evidence from Japan, South Korea, the United States, and beyond suggests there is no economic downside to flexible copyright regimes – quite the opposite.
4) Copyright Solutions Are Not Cost-Free
Obliging relicensing payments for AI training comes with significant financial and other transactional overheads. Any relicensing scheme will depend on the efficiency and good governance of the organisation managing it, at a time when the EU is still working to ensure full implementation of the Collective Rights Management Directive. Such costs are likely to be particularly harmful for all but the largest players, and in particular the research sector. Furthermore, licensing solutions are far from comprehensive, with inevitable opt-outs and complexities arising where intermediaries don’t necessarily have the rights, or if they do, not necessarily in the jurisdictions needed.
5) Licensing Promotes Bias and Poor AI Predictions
The quality of AI is determined by the three ‘Vs”– velocity, volume, and veracity. It is a matter of fact that, where appropriately processed, greater access to information (i.e. volume and veracity of content) leads to better quality AI less prone to bias. Conversely, restrictions on the availability of content, as a rule, will increase the chance of bias and poor model predictions. Given that licensing restricts the volume and veracity of what a model trains on, models developed in Europe will inevitably be less accurate in their predictive power compared to those developed in North America and Asia, which use flexible models to allow for training on legitimately acquired materials. This is something that should attract far more attention from policy makers than it currently does.
6) Smoke and Mirrors: AI is More Than Generative AI
One problem in the debate has been the way in which AI itself is presented. AI companies have, of course, painted a picture of highly sophisticated and capable systems, boasting huge valuations despite being far from profitable. Meanwhile, commentators all too often conflate generative AI with all AI, even though most models produce no generative outputs at all. Even within generative AI, there is confusion between those outputs that may be relevant to cultural markets with others that are not. A generative model helping with diagnostics has nothing to do with a model that can produce artwork.
7) Inflexible Copyright Rules Make Europe Uncompetitive and Benefit Market Incumbents
Given the unfriendliness of the environment, comparatively little training using copyrighted materials is taking place in Europe today. Because of copyright, players that can train their models outside of Europe place their offices and servers in the US and Asia. Where models require information to be sourced from the EU, or for processing to take place here, the complex copyright rules around AI only benefit the market incumbents with the deepest pockets that have the most chance of securing large numbers of licences for AI. As a second-comer in AI markets, the EU should be doing the opposite – making access to data easier, not harder for start-ups, SMEs, and research-oriented organisations of all types.
There is, however, still potential for the future. To realise this, there needs to be less focus on rules that are likely to have little impact due to “jurisdiction shopping” other than to chase research activities to other parts of the world, and more on how to enable rather than chill new ideas and innovation.
8) Europe’s Research Ambitions Suffer From a Copyright Blindspot
Ironically, while one corner of the debate on AI is overly-fixated on copyright as a tool, more broadly there appears to be a fear to touch it. This may be due to the fear of noisy responses from the copyright industries, but ultimately this risks accepting that it is simply more complicated and expensive to invest in AI in Europe. As pointed out above, this simply makes Europe unattractive and will lead to lower quality (and less safe) outputs. This seems very far from the ambition that international researchers and the research-intensive industries will ‘Choose Europe’.
9) Smart, Not Dumb, Regulation
It is naive to imagine a horizontal body of law like copyright can address and support the different requirements raised by the research-intensive sector, startups, scaleups, SMEs, the creative industry, etc, across something so very broadly deployed as AI. All the arguments made point to the need for a different type of regulatory agenda around AI that is realistic about the challenges, deals with issues sector by sector, but not dogmatic about the solutions.
In addition, we should not forget that there are plenty of policy tools already available. Existing rules around copying are enough to make it illegal to use AI outputs that too closely resemble copyrighted works, or at least to oblige the payment of royalties. Meanwhile, there are well-established rules around privacy and discrimination, competition tools designed to avoid major actors harming wider markets, and of course well-established approaches to ensuring ethical practice in research.
Crucially, it will be important to focus on actual harms—for example protecting rightholders’ ability to exploit materials commercially, but not hold back other uses across the 96% of the economy.
10) A Targeted Approach to Supporting Creativity
Outside of regulation, the challenge of falling creator revenues long pre-dates AI, and points to the need for focused cultural policies. Taxing artistic generative AI profits or platforms, or imposing copying levies could offer a way of raising funding—alongside others—for more targeted support for creators. Such targeted approaches could also help counteract the redistribution of incomes towards the largest players in the entertainment sector that copyright solutions offer, given how the royalties ending up in the hands of most creators are likely to be.
17 September 2025
