Everyone is celebrating the French for defending their magnificent culture through a law that would force AI services to face the music. The bill, which complements the implementation of the EU AI Act, creates a reversed burden of proof in copyright disputes, meaning that an AI provider must demonstrate that its system did not use a protected work or that any use was lawful, rather than requiring the rightsholder to prove infringement. This exposes AI developers to a higher likelihood of injunctions and damages claims because courts can act on the presumption of use unless the provider can rebut it with credible evidence. As a result, developers must maintain full dataset traceability, robust documentation, and audit-ready evidence showing how training data was sourced, processed, and verified. Rightsholders gain a significantly stronger litigation position.
But while it is an important first step, as the lawyers would say: hold your horses. The French Senate has adopted this bill on copyright liability and evidentiary rules—originally proposed in December 2025—with several amendments. It now goes to the Assemblée nationale for further discussion and likely amendments. The two-chamber process is similar to the UK: both houses must agree on an identical text. Instead of “ping-pong”, France uses the Commission mixte paritaire to negotiate a compromise when the chambers disagree. It sounds more sophisticated.
Given the broad political support—unanimous in the Senate—and the usual vive la créativité sentiment in France, the bill could move quickly, potentially even before July.
The Senate has discussed but not yet created a collective management system for AI training, but the political, legal, and practical momentum is clearly moving in that direction. The presumption of use is widely seen as the structural precursor that makes collective licensing the inevitable next step — liberté, fraternité, collectivité. As always, qui vivra verra.
Florian Koempel (International Consultant)