Issue #17 - April 2, 2026
Is EU Inc. meeting Draghi's threshold? And the AI copyright levy debate returns
Sébastien Louradour
4/2/20264 min read
With EU Inc., has the EU met Draghi’s Digital Single Market threshold?
On March 18, the European Commission introduced EU Inc., an initiative that leading European startups have advocated for for many months now and that was initially presented as the 28th regime in Letta’s report. The original goal was to create a single regime for startups, virtually operating in a 28th State, providing simplified rules to scale across all member states. The argument that at the time made the idea so compelling is that it is easier for a European startup to scale in the US than in the EU. The proposal suggested by the EU Commission is according to startup advocacy groups a good step forward, but with still a lot to be done to have a similar regime than the US one. On the bright side, EU Inc introduces faster incorporation, digital-first processes, and stronger tools for investors and talent. Among them, the capacity to create the company in 48 hours and for 100 euros, but also a harmonised tax regime that would allow owners of stock to be taxed only when they sell. Unsurprisingly, the underwhelming part of the proposal lies in the difficulty for member states to give up on remits that have consistently been a national competence. A fully harmonised model will be very hard to implement, with corporate tax rules, VAT, labour laws, social security that will stay national. There is nothing new here, the EU has gone through treaties that have over the past decades extended the role of the EU over member states but to a certain extent. To fully materialise and meet its initial ambition, EU inc. will require a broad range of ambitious reforms that go far beyond the harmonisation of the startup regime. Many tech organisations have gathered around EU Inc Org to advocate for this regime, and they have been quite successful at driving the conversation so far. They are already calling for a more ambitious text, specifically focusing on the lack of predictability of the regime with the risk of having to deal with 27 courts of justice, with varying interpretations, instead of a single one for all EU-Incs. The Country of origin principle put in place for GDPR enforcement has recently proven its limits, and while the EC is working on a revision of the harmonisation of enforcement digital rules, the EU has an opportunity this time to succeed from past experiences. Still, It will require to strike a delicate balance of influence between member states and the EU. The EC has nonetheless expressed willingness to go fast on this file, and it’s likely the Irish Presidency of the EU covering the second semester of 2026 will take over the file and push it through the trilogues for a potential adoption by the end of 2026.
Debate over right-holder treatment for AI model training is back
On March 20, The founders of Mistral, among them its CEO Arthur Mensch, published a widely noted op-ed in the FT calling for a new levy on AI labs to compensate copyright holders. This letter marks a new development in a long series of back and forth between right-holders and AI labs on how protected content should be considered at the age of GenAI, and most likely, a response to recent legislative developments on the matter in France. In December, a few French senators proposed a law to update the French intellectual property law that acted as a bombshell among AI Labs. The goal of the law proposal is to create a presumption of copyright data use for AI model training. This proposal, for which France's highest administrative court, the Conseil d'État, issued a favourable advisory opinion, would represent a complete change of direction on how copyright data is currently being approached in the EU. The “opt-out regime” that was adopted by the EU with the copyright directive has been a delicate act of balancing from the start but was ultimately voted on and transposed by each member state. While the opt-out regime allows right-holders to ask for their protected content not being used for model training, the proposal made by the French senators would go much further by considering that all labs are by default using protected content and that they should prove it’s not the case. The debate over copyright started dramatically over the AI Act Code of Practice discussion when right-holders started to wonder whether protected content could either be more protected or be part of commercial agreement with AI Labs over supposedly uncontrolled use of internet data scraped for model training. Still, a status quo had been reached, and a majority of AI Labs had signed the Code of Practice. Mistral’s proposal in the FT suggests another and more upfront way of addressing the core issue: right holders need a levy to survive the AI revolution. The Senate is due to examine the law proposal in plenary on April 8, and if adopted by the high room, The National Assembly would then need to confirm the initial vote, opening up to a new series of political debates. While the adoption of such law would represent an historic victory for right-holders, it would cool down the appetite of investors and AI Labs betting on France for its large pool of top engineers and nuclear power plants able to boost data centers. In this context, Mistral’s proposal may be seen as a more reasonable approach, and a potential credible alternative to the current Senate proposal, by creating a more predictable copyright regime for AI Labs while acknowledging that protected data provide more quality and value. There is undoubtedly something bigger playing here too, with Mistral offering an alternative to dominant US platforms, it can foresee an opportunity to influence the course of tech development and promote its own vision and values on the future of AI. France has always praised its “cultural exception” and in this grand vision for the role of culture, the most en vogue French startup seems to have decided to lean towards Diderot and Rousseau too.