Between Arthur Mensch and the world of creation, reconciliation is not yet on the horizon.
In the background, a bill aimed at establishing a “presumption of use of cultural content” by AI providers. It is the work of Senators Laure Darcos (Essonne, group Les Indépendants – République et Territoires), Agnès Evren (Paris, LR) and Pierre Ouzoulias (Hauts-de-Seine, PCF).
Mistral Favors a “European Creation Fund”
The text was filed in December 2025. On 8 April 2026, the Senate passed it, almost in its initial wording. The main provision: there is a presumption that an AI system has exploited a work as soon as an indicator related to the development or deployment of the system or to the result generated by it makes such exploitation plausible.
A few weeks earlier, Arthur Mensch had suggested an alternative path. Specifically, an “European Creation Fund.” AI model providers would contribute to it in proportion to their revenue.
The chief executive of Mistral clarified that this mechanism would not replace licensing agreements. And more broadly the “direct relationships” between creators and AI providers. The scheme would nonetheless offer a legal certainty, contrasted with the opt-out framework that is “impossible to put into practice.”
The Publishers Association (United Kingdom) had reacted. “Arthur Mensch assumes that tech companies are entitled to use whatever they want without permission and to ask forgiveness later,” said its chief executive. And urged Mistral to settle commercial licenses, “just as it pays its employees or its electricity bills.”
Rights Holders Demand Licensing Agreements
The National Assembly was slated to discuss the text in a public session in June. The 110 amendments filed contributed to pushing the examination to 19 November.
As the deadline approached, creators raised their voices again. Representatives from culture and information (SACEM, SNE, SACD, etc.) co-authored a tribune. “France does not have to arbitrate between technological prowess and cultural power,” they assert. If credible, AI providers do not have to fear a proliferation of lawsuits, nor an infringement of trade secrets. A response to Arthur Mensch, who says he fears both.
The accused laments that a presumption of use triggered by a mere inverse index would reverse the burden of proof. According to him, proving that a work was not used would require, across a corpus of billions of words, exposing industrial secrets. Acknowledging that the presumption is rebuttable and limited to civil law, he nonetheless insists: at the scale of training a frontier model, proving the negative without revealing its methods is impossible. The result is written in advance: an inextricable juridification of the relations between rights holders and AI actors. Until a European text comes along to replace this law anyway.
“Judges Will Organize Confidentiality”
“You don’t want any regulation in France, while the Americans themselves are asking questions,” retorts Laure Darcos. Trade secrets cannot excuse a violation of the fundamental property right, the senator continues. Not without declaring that “the judges will organize confidentiality.”
Same tone from her colleague Laurent Lafon (Union des Centrists & Independents, Val-de-Marne): “Arthur Mensch is waging a backward-looking battle.” The lawmaker cites California, which has begun legislating to frame AI development. “For ethical and economic reasons, we have an interest in shaping this regulation ourselves,” he proclaims.
At the ARP Film Encounters (Society of authors, directors and producers), Xavier Bertrand voiced support for the proposed bill. “The danger is real: with AI development, we risk seeing our creators deprived of their works and, tomorrow, our imagined world crafted elsewhere,” commented the president of the Hauts-de-France region.
Following Laure Darcos, Isabelle Degeorges notes that judges “know how to safeguard trade secret confidentiality.” In keeping with her peers, the president of Gaumont Télévision is opposed to Arthur Mensch’s European Creation Fund proposal. “Creators are not beggars. We demand that our rights be respected, that the works used are disclosed, that licenses can be negotiated.”
According to Blaise Mistler, head of institutional relations at SACEM, rights holders have “done their part.” On one hand, by engaging in talks with Mistral (“without achieving real licensing negotiations”). On the other, by proposing to refocus the mechanism on generative AIs that produce cultural and informational content comparable to protected works.
Renaissance Opposes the Text…
In the National Assembly, Éric Bothorel also tried, from the committee stage, to modify the bill’s scope. The deputy for Côtes-d’Armor—affiliated with Ensemble pour la République, formerly Renaissance—indeed filed an amendment “worked with the French Telecommunications Federation” to replace “system” with “model.” And thus exclude downstream actors. Rationale: “An application scope that is too broad and unjustified […] would stall the growth of the sector and our digital sovereignty.”
The Committee on Cultural Affairs, Education and Sport rejected this amendment. Just as it rejected many others that Bothorel had filed. One of them was simply a motion to delete. It echoed the arguments presented by the National Bar Council. First, the existence of a European legal framework already harmonized and fully applicable. In its crosshairs are the DAMUN Directive (2019/790) and the AI Act. The former provides for text and data mining exceptions with a clear right of opposition for rights holders. The latter imposes transparency obligations on training data for general-purpose AI models.
… and Seeks to Narrow Its Reach
Éric Bothorel argues that the proposed law departs from that harmonized framework: it alters the logic of the European rights of the actor, which rests on proof of reproduction or a public communication. The lawmaker maintains that the mechanism creates legal uncertainty: the criterion of an “index making it plausible” is “broad and imprecise,” opening the door to lawsuits based on indirect elements (prompts or user practices, statistical or stylistic similarities…). He also warns of a risk to the presumption of innocence and to the right to a fair trial.
Other amendments from the deputy for Côtes-d’Armor aimed to:
- Condition the presumption on prior opposition by the rights holder
- Limit it to a finite and cumulative list of circumstances
- Not allow it to be tied to the “generated result”
- Not apply it to a provider that complies with the AI Act’s transparency obligations
- Exclude works that have fallen into the public domain
The Ensemble pour la République group had filed further amendments. In particular through Prisca Thevenot (Hauts-de-Seine). She is at the origin of most of the 110 amendments that were to be examined in the public session.
