It’s a Trap: Why Strengthening Copyright to Protect Creators Will Do The Opposite
Admiral Ackbar, the rebel commander from Return of the Jedi, is remembered for exactly one line: “It’s a trap.” The European Commission’s call for evidence on the CDSM Directive review, open until 25 June 2026, deserves the same warning.
Two recent analyses, one legal, one economic, converge on the same conclusion: reaching for copyright to manage the disruption caused by generative AI is a misdiagnosis. The legal framework was not built for this problem. The economic consequences of forcing it to do the job anyway fall hardest on the smallest players in the digital economy. Policymakers walking into this consultation need to understand what they are actually being asked to do.
A right designed for a different fight
The press publishers’ right, codified in Article 15 of the CDSM Directive, was designed to rebalance power between news organisations and search engines. It was already struggling to do that job before generative AI arrived. Legal scholar Viktoria Kraetzig, writing in the International Review of Intellectual Property and Competition Law, argues that the right suffers from a foundational design flaw: it is a property instrument deployed against a market-structure problem.
The logic of Article 15 assumes that the leverage of an exclusive right will force information society service providers to negotiate and pay. That assumption held only weakly in the search engine era, where publishers and platforms existed in a symbiotic relationship: platforms depended on content, publishers depended on referral traffic. Neither side could afford to walk away. The result was a standoff, not a settlement.
AI breaks the symmetry entirely. The training of a large language model does not require a continuing relationship with the publisher whose content contributed to it. There is no traffic to redirect, no audience to lose. The content has already served its purpose before any negotiation can begin. The exclusive right over reproduction and making-available, the two levers Article 15 provides, cannot reach back into that process.
The territorial limit compounds this. AI training takes place overwhelmingly in the United States and China. EU copyright law does not follow data to San Francisco or Beijing. The press publishers’ right, as Kraetzig concludes, reaches its expiry date precisely at the moment the creative industries need it most.
The deeper problem is structural. Kraetzig argues that the right regulates the relationship between press publishers and the information market as it existed in 2019, not as it exists in 2026. Successful publishers, The Economist, the New York Times, Die Zeit, have adapted. They offer something AI cannot replicate: verified, contextualised, editorially accountable journalism that readers will pay for directly. The titles in distress are those whose product was already undifferentiated before AI arrived. Copyright law cannot restore competitive advantage that was never rooted in copyright.
A trap with three entry points
The argument that copyright is the wrong tool is not new. Legal scholar Carys Craig, a former guest of our podcast, identified the architecture of the problem in a paper published in the Chicago-Kent Law Review, characterising it as the AI-copyright trap: the mistaken conviction that copyright law is the best instrument to support human creators in the current technological moment, when it is likely to do more harm than good.
Craig identifies three routes into the trap. The first is the assumption that if something has value, someone must own it. This logic produces the conclusion that AI companies are stealing from creators simply by training on their work, even where no protected expression is reproduced in the output. The second is the assumption that copying is inherently wrongful. In the digital environment, every technical process involves copying. AI training involves the extraction of statistical patterns from data. The copy is incidental to the learning; treating it as the unit of analysis misdirects the regulatory response. The third route is the belief that stronger copyright protection will translate into income for working artists. History does not support this. Publishers, collecting societies, and rights aggregators have consistently captured the economic benefit of copyright expansion before it reaches individual creators.
The trap is effective because each of these routes is emotionally compelling. Creators are being displaced. The companies training on their work are enormously profitable. The demand for consent, credit, and compensation resonates. But as Craig demonstrates, acting on that demand through copyright mechanisms is likely to produce rules that serve media conglomerates and collecting societies far better than they serve the freelance illustrator or the independent journalist.
The economy that gets caught in the net
ECIPE’s “The Copyright Trap” policy brief introduces the dimension that political discourse consistently ignores: the cost to the rest of the European economy. The creative industries generate approximately 202 billion euros in value added. The data-intensive sectors of the wider economy, ICT, financial services, automotive, pharmaceuticals, chemicals, are substantially larger. ECIPE calculates that offsetting a one percent reduction in value added across those sectors would require the creative industries to grow their output by more than nine percent.
That asymmetry matters because the text and data mining exception in Articles 3 and 4 of the CDSM Directive is not just a provision relevant to media companies and AI developers. It is a foundational input for the entire European industrial economy. Autonomous vehicle systems cannot be trained on simulation alone. Drug discovery models require diverse biological datasets. Climate forecasting tools need continuous retraining on remote sensing data. Fraud detection systems in European financial services must adapt to shifting transaction patterns. All of these applications depend on broad, legally stable access to training data.
The EU has invested heavily on one side of this equation. The EuroHPC AI factories, the InvestAI programme targeting 200 billion euros in mobilisation, the Apply AI Strategy directing one billion euros into ten strategic sectors: these are expressions of a political commitment to European AI competitiveness. On the other side, the ECIPE analysis documents a paradox: every regulation that narrows the TDM exception acts as a levy on AI development across the entire economy, not just on the companies the regulation is targeting.
The comparison with competitor jurisdictions makes the stakes concrete. Japan’s Article 30-4 permits the use of copyrighted works for information analysis regardless of commercial purpose, provided the work’s expressive content is not enjoyed. Singapore’s computational data analysis exception explicitly overrides contractual restrictions, providing developers with legal certainty that no amount of bilateral negotiation can replicate. The EU’s TDM framework, already conditional on the opt-out mechanism in Article 4(3), is not exceptional by global standards. It is merely less restrictive than some alternatives. Narrowing it further would shift AI development to jurisdictions where no equivalent restriction applies, taking the tax revenue and employment effects with it.
Who actually bears the cost
The political argument for tighter copyright rules presents itself as a defence of the small against the large: independent creators against trillion-dollar technology companies. The economic reality runs in the opposite direction.
Large AI developers have the legal teams to navigate a fragmented opt-out landscape and the balance sheets to negotiate bilateral licensing agreements with major content providers. Axel Springer, News Corp, and the Financial Times have already moved in this direction, signing direct deals with OpenAI and Mistral. This is not a market failure requiring regulatory correction. It is the market functioning as markets do: parties with bargaining power reaching agreements.
Mistral co-founder and CEO Arthur Mensch made this point with unusual directness when testifying before the French National Assembly’s commission on structural dependencies on 12 May 2026. Describing the cumulative burden of the GDPR, the AI Act, and copyright rules, he told deputies that the regulatory stack was manageable for Mistral because the company had reached sufficient scale. The implication was not reassuring: regulation favours the large. Younger companies, unable to absorb the compliance cost, are choosing the United States instead. Europe is not losing them to superior US technology. It is losing them to a lighter administrative burden. Mensch’s framing of AI as strategic infrastructure comparable to energy makes the stakes concrete: if the companies building European models cannot access training data on legally stable terms, the infrastructure does not get built in Europe. It gets built somewhere else, and the economic value follows.
The same asymmetry applies within the creative industries. The interests articulated loudest in Brussels are those of large collecting societies and major publishers, who stand to capture the administrative fees and a substantial portion of any licensing revenue that a mandatory regime would generate. Individual authors, photographers, and journalists are presented as the beneficiaries of the proposed rules. They are more likely to be its nominal justification.
What the consultation should actually address
The Commission’s call for evidence is structured around a set of real concerns. Creators are losing income. The licensing market is fragmented. Opt-out mechanisms have not yet been fully adopted in practice. Performers face AI-generated imitations of their voices and likenesses. These are actual problems, but they are not all copyright problems.
The displacement of human creative labour by AI-generated substitutes is an economic and labour market problem. It is analogous to the displacement of theatre by television, or of physical music retail by streaming. Neither of those transitions was resolved by restricting the enabling technology at the input stage. Both were managed through a combination of market adaptation, new business models, and targeted public intervention such as arts subsidies and cultural funds. The creative industries have navigated technology shocks before. The policy question is how to support that navigation, not how to halt the technology.
The question of performer imitation, AI-generated voice and likeness replicas, does not fit neatly within the CDSM framework at all. It is a personality rights and unfair competition problem. Kraetzig’s analysis suggests the more appropriate regulatory frame is not property but fairness: whether conduct between competitors on the information market exceeds a threshold of acceptable imitation. That question has existing legal infrastructure, in unfair competition law and in the emerging framework around data ethics, that is better suited to the specific harm.
On the core TDM question, ECIPE’s recommendation is clear: treat the existing exception as a strategic asset, not a concession. Subject any proposal to narrow it to a mandatory competitiveness impact assessment. Resist the expansion of the opt-out mechanism. Allow the private licensing market, which is already producing results, to develop without being pre-empted by a statutory regime that would lock in current market structures.
The trap is about to close
The 25 June deadline for the Commission’s call for evidence will produce a record dominated by rights-holder submissions. That is the structural reality of European copyright consultations. The creative industries are organised, articulate, and motivated. The diffuse economic interests of downstream data users, the automotive engineers, the pharmaceutical researchers, the climate modellers, the European AI start-ups, are harder to aggregate into a coherent lobbying position.
The academic and economic literature is less ambiguous. Two independent lines of analysis, one from copyright doctrine, one from industrial economics, arrive at the same place: using copyright to manage the economic disruption caused by generative AI is the wrong intervention. It will not protect creators in any meaningful sense. It will concentrate power in the hands of incumbent intermediaries. It will impose a data scarcity tax on the sectors of the European economy that most need access to abundant training data. And it will push AI development toward jurisdictions that have made a different and more rational regulatory choice.
The copyright trap is not a metaphor. It is a description of a mechanism. The bait is compelling. The mechanism is already in motion. The Commission’s review is the moment to examine whether the trap should be allowed to close.
Written by Caroline De Cock, LL.M., Head of Research
