French bill would presume AI copyright infringement from a single clue, warns Mistral CEO
eliebakouch · x · 2026-11-08
French politicians are pushing a bill where a single "clue" — AI output resembling an author's work — triggers a presumption that the work was used in training, reversing the burden of proof onto model providers.
Mistral CEO Arthur Mensch responded: while compensating creators is legitimate, proving a negative across billions of words is impossible without exposing industrial secrets. He argues the legal risk will fall first on French players at a critical moment for decade-scale infrastructure decisions, and proposes instead a levy on AI providers' turnover in Europe feeding a creation fund (aligned with AI Act Article 53). He calls on lawmakers to reject the text, warning that otherwise American and Chinese players will dictate export conditions.
Related event: French Bill Would Presume AI Copyright Infringement From Output Similarity(2 posts)→
More from Safety
- SvD maps 3 AI catastrophe scenarios with 50 experts — anderssandberg · 2026-11-08
- CS student shares permission design principles for agents that touch real files — ASK-1T · 2026-11-08
- Mistral CEO opposes French copyright bill while building 1GW of European compute — arthurmensch · 2026-11-08
- iOS keyboard's hidden blocklist cracked: 4,500+ secretly banned swipe phrases exposed — TimDarcet · 2026-11-08
- Open-source tool Obiliteratus claims to uncensor any AI model from your browser — udmrzn · 2026-11-08
- Stylo now auto-scans a LLM-focused subreddit and bans AI-written posts — FeydRowan · 2026-11-08