Europe’s AI Ambitions vs. the Voss Report: A Collision Course

Despite calls to reject the document (by among others COMMUNIA and EARE letter signed by 116 individuals and organisations) the JURI Committee decided to adopt the controversial own-initiative report on Copyright and Generative AI, penned by none other than CDSM rapporteur, MEP Axel Voss. 

As we have stated before, the ideas expressed in this report represent a profound strategic misalignment with the goals of advancing AI creation in Europe and regaining our failing competitiveness. While the European Commission’s “Apply AI” Strategy and the “AI Continent” Action Plan aim to “accelerate AI development” and “leverage the Digital Single Market,” this report calls for drastic regulatory interventions that, if adopted, will put European AI creators at an acute disadvantaged and stall European innovation at the starting line.

While some improvements were introduced to the report through compromise amendments, and some of the most egregious ideas and misrepresentations have been removed, it is still a deeply misguided and destructive set of anti-innovative rhetoric. 

The glaringly false notion that generative AI is not covered by the exceptions in Articles 3 and 4 CDSM was fortunately removed. The report however still calls the application of Text and Data Mining (TDM) exceptions to Generative AI as “ambiguous,”creating uncertainty and legal risk.  This remains in direct contradiction to the AI Act’s Recital 105, which explicitly anchors AI training within the existing TDM framework, and the recently adopted Code of Practice on Generative AI, with its extensive chapter on copyright obligations. 

The report also call for a de facto removal of press publisher content from the scope of the TDM exceptions, thus creating a toxic two-tier content regulation structure, which will without a doubt create endless legal disputes on what the definitions and scope, as well as possibly hamstringing all AI systems in Europe that would be cut off from the grounding necessary to avoid hallucinations and deliver up-to date responses. This would introduce another exclusive right for press publishers, giving them veto power over AI training and deployment in the EU. 

Another huge danger posed by the report is the call for an “itemised list identifying each copyright-protected content used” which, as anyone with a basic understanding of the scale of AI training, is a technical impossibility for Large Language Models (LLMs) trained on petabytes of data. This requirement creates a compliance burden that only huge non-European incumbents can even think of lifting, directly opposing the EC’s goal of “fostering a domestic AI ecosystem.” This also directly contradicts the AI Act and the Code of Practice, which demands a “sufficiently detailed summary” of the data used. This is already a serious disadvantage for EU companies, but at least it is physically possible to meet. The fact, that the “presumption of use” (ie. the susceptibility to being sued by rightsholders) was changed to “rebuttable” is a small consolation in this context. 

And in case you were hoping this would only apply to foreign hyperscalers and providers of frontier models the report expressly pulls deployers of AI systems into its scope. In particular activities such as Retrieval-Augmented Generation (RAG) and fine-tuning are singled out as triggering copyright liability. This is the bread-and butter of thousands of EU companies, most of them SMEs. There is no plausible deniability of this being a direct attack on the entire EU AI ecosystem. 

All this runs directly counter to the stated goals of the EC and national governments, that seem to have understood how important the fostering of a home-grown AI creator ecosystem is for both direct and indirect economic growth, but perhaps more importantly for maintaining European technological sovereignty. 

The Commission’s AI Continent Action Plan and Apply AI Strategy emphasizes the need for a “predictable legal environment” to encourage SME investment. However, the Voss Report actively undermines this by calling for the introduction of more legal risk and friction for EU based entities. 

Meanwhile elsewhere in the world the US Fair Use doctrine (17 U.S.C. § 107) remains a flexible, technology-neutral tool that supports transformative uses like AI training, as seen in ongoing jurisprudence and Article 30-4 of the Japanese Copyright Act provides one of the world’s most permissive environments for TDM, allowing for the use of copyrighted works for “information analysis” regardless of the medium.

If the European Parliament intends to uphold the AI Continent vision, the EP Plenary must reject the overreaching provisions on press media and the technically illiterate transparency mandates. We cannot lead the “AI Revolution” if we are the only continent that makes it a legal liability to read the internet.