EU copyright loophole in AI Act leaves creators exposed to Big Tech

A legal gap in the European Union's Artificial Intelligence Act has left writers, musicians and other creators unable to enforce copyright against generative AI systems trained on their work, according to the MEP who helped write the bloc's 2019 copyright directive. Axel Voss, a German centre-right member of the European Parliament, says the legislation was never designed to handle models that ingest vast troves of books, articles, images and songs to produce synthetic content.

The exemption that became a loophole

At the centre of the dispute sits the text and data mining (TDM) exemption in the 2019 Copyright Directive. Article 4 permits reproduction of lawfully accessible works for TDM purposes unless rights holders have expressly reserved their rights in an appropriate manner. When the directive was negotiated, Voss says, the exemption was conceived for academic researchers and limited private use, not for the world's largest technology companies to harvest intellectual property at industrial scale.

The AI Act, which entered into force in August 2024, requires providers of general-purpose AI models to comply with the 2019 directive. But Voss argues this cross-reference creates a misunderstanding: the TDM exemption cannot legitimately cover the training of generative AI models on published creative works. A 2024 academic study by legal scholar Tim Dornis and computer scientist Sebastian Stober reached the same conclusion, finding that such training constitutes copyright infringement rather than text and data mining.

"The introduction of the TDM exemption in the AI Act was a misunderstanding," Voss said. He added that he failed to secure majorities among EU lawmakers for stronger copyright protections during the AI Act's final negotiation stages. The absence of such provisions was "irresponsible" and it was "unbelievable" that the gap remained.

Creative industries sound the alarm

The practical consequences are already visible. Nina George, a German bestselling author whose works have been translated into 37 languages and who serves as president of honour at the European Writers Council, described the TDM exception as "devastating". Copyright exclusions, she said, were originally intended to balance authors' interests against the public good, allowing schools to photocopy texts, for instance. "These AI exceptions for commercial use mean that business interest will be served for the first time," she said. "This is a shift of paradigms and a perverted way to bend copyrights and authors' rights to serve the interest of a few businesses."

George said she has no way of discovering whether her books have been used to train generative AI systems. "The lack of instruments to enforce any rights, this is the scandal in the construction of the AI Act in relation to copyright directive." Aafke Romeijn, a Dutch-language electropop artist and board member of the European Composer and Songwriter Alliance, echoed the point: there is no practical opt-out mechanism for creators whose work feeds AI applications.

The European Council of Literary Translators' Associations, representing 10,000 translators across 28 countries, stated that books are written by human authors and must be translated by human translators to preserve artistic virtues. The organisation insists that authors, performers and creative workers must have the right to decide whether their works can be used by generative AI and, if they consent, to be fairly remunerated.

Transparency requirements fall short

The AI Act does impose a transparency obligation. From 2 August 2025, providers of general-purpose AI models must publish a sufficiently detailed summary of the content used for training. But the devil lies in the implementing details. Voss said the latest draft rules from the EU's AI Office are "not sufficiently detailed" to protect artists. In a letter to the Commission this week, 15 cultural organisations warned that the draft summary proposals fail to ensure meaningful transparency. More broadly, they wrote: "The impact of AI on the authors and performers we represent constitutes a systemic risk."

Romeijn said senior EU officials have told creators to take technology companies to court to preserve their copyright. "Who is actually going to take a big tech company to court?" she asked, citing cost, time, loss of earnings and potential reputational damage. "It is just a very impractical way of implementing legislation."

Political division over the Act's strength

Brando Benifei, the Italian Social Democrat who jointly led the European Parliament's negotiating team on the AI Act, contests the view that creators are unprotected. He described the legislation as "a very strong text" with the potential to create "a very large rebalancing of power between the developers and the rights holders". But he acknowledged that from "day one" after the law was adopted, there had been an effort "to dilute and to interpret in a minimalistic way the provisions". This, he said, "has been the obsession of the big tech companies because it is probably the part of the AI Act that can be most impactful in terms of costs for the big generative AI companies."

The disagreement between Voss and Benifei reflects a deeper tension. Both were involved in shaping the AI Act, but they sit in different political groups and read the final compromise differently. Voss sees a structural flaw; Benifei sees a strong framework under attack from lobbyists. The Commission, for its part, says it is "closely monitoring the global challenges that AI technology development poses to the creative industry" and remains "committed to maintaining a balanced approach that fosters innovation while protecting human creativity". A spokesperson added that the executive is "assessing the need for additional measures, outside the AI framework", declining to say whether this means new legislation.

The missing liability directive

Voss had hoped that the proposed AI Liability Directive, which would have eased the burden of proof for claimants alleging harm from AI systems, might fill the enforcement gap. But the Commission withdrew that proposal last week as part of its 2025 work programme simplification drive. "It might be getting very difficult," Voss said. "And so the infringement of copyright is continuing, but nobody can prove it."

The withdrawal removes a potential legislative vehicle for addressing the evidentiary problem that creators face: without transparency about training data, they cannot establish that their works were used, let alone secure compensation. The Commission's decision to drop the liability directive while the AI Act's copyright provisions remain contested has reinforced the perception among cultural organisations that the executive is prioritising tech sector competitiveness over creative rights.

Silence from the Commission

In December 2024, most of the same 15 cultural organisations wrote to Henna Virkkunen, the Commission's executive vice-president for tech sovereignty, security and democracy, raising concerns that EU law "fails to adequately protect the rights of our creative communities and the value of their cultural works". Nearly 11 weeks later, according to three signatories, the Commission had not replied. "So far it does not seem that she has an ear or an understanding, I am sorry to say that, of the whole value chain and how it works in the cultural and creative industries," George said.

The silence is notable. Virkkunen's portfolio includes the Digital Services Act, the Digital Markets Act and the AI Act, the three pillars of the EU's digital regulatory architecture. Her failure to engage with a detailed letter from representative bodies across the creative sector suggests either institutional bandwidth constraints or a political judgment that the current framework is sufficient. Neither interpretation reassures the signatories.