Culture Minister Rachida Dati has put the issue on the table: France is considering a specific law to protect creators and cultural industries against the massive training of artificial intelligences on their protected works.
Mandatory remuneration for the use of protected content?
The stated objective is clear: to allow publishers, authors, photographers, musicians, and media to negotiate, or directly obtain, remuneration when an AI like ChatGPT (OpenAI), Gemini (Google), Llama (Meta), or other models train on their books, articles, photos, or music tracks.
The ministry estimates that millions of French works have already been absorbed without authorization or financial compensation.
Reversal of the burden of proof: AIs will have to prove they are clean!
Among the most audacious avenues put forward by Rachida Dati is the “reversal of the burden of proof.”
Currently, it is up to the author or publisher to prove that an AI has used their work. With the future law, it would be up to AI companies to demonstrate that they have not exploited protected content, or that they possess the necessary rights.
A “presumption of use” would be established: it would be assumed that any large-scale model has necessarily ingested a significant portion of the French and European web.
A timeline that raises questions
Several voices in the cultural sector point out that the initiative comes late. Current large models (GPT-4, Gemini Ultra, Claude 3, etc.) have already been trained for years on colossal datasets very likely including millions of French works.
A future law will not allow for a reversal of this past training, but it could impose strict rules for updates and future models.
Read also: Will Gemini 3.0 Pro eclipse GPT-5.1 in the frantic AI race?
Towards a European exception or a tougher French model?
While the European Union is already working on the AI Act and data transparency for training, France seems to want to go further and faster, particularly on the issue of remuneration. Rachida Dati has repeatedly stated that American AI giants make colossal profits from content created by others, often without any compensation.
The ministry is therefore studying a national mechanism that could serve as a laboratory before eventual European harmonization. A draft law could be presented as early as 2026, although no precise date has yet been communicated. Discussions with rights holders, publishers, and AI platforms promise to be lively.
Source: LeMonde
