Policy

Darcos Bill That Would Make AI Firms Prove They Did Not Use Protected Works Heads for a Possible November 19 Assembly Vote

· 6 min read · Policy

Darcos Bill That Would Make AI Firms Prove They Did Not Use Protected Works Heads for a Possible November 19 Assembly Vote

A French bill that would shift the burden of proof onto artificial intelligence providers in copyright disputes could come to a vote in the National Assembly on November 19, Les Echos reported on November 14, describing a last-ditch lobbying battle between the tech and cultural sectors. The bill from Senator Laure Darcos (Horizons) would create a presumption that protected works have been used by AI models: where there is evidence making such use plausible, it would be up to the provider to show otherwise in civil proceedings.

The text was adopted unanimously by the Senate on 8 April 2026 and approved by the Assembly’s cultural affairs committee on 2 June, according to Reporters Without Borders (RSF), which supports it. It was first placed on the agenda of the Democratic and Republican Left group’s reserved day on 11 June, but too low down to be debated, after the presidential Ensemble pour la République (EPR) group tabled 100 of the 110 amendments filed on the single-article text. The Socialist group then put it on its own reserved day of 19 November, initially in sixth place. RSF has called on Gabriel Attal, who leads the EPR group and is a declared candidate in the 2027 presidential election, to withdraw the amendments.

What it means in practice

Darcos Bill That Would Make AI Firms Prove They Did Not Use Protected Works Heads for a Possible November 19 Assembly Vote — contextual photo

Les Echos reports that organisations defending authors and producers have worked on a compromise amendment meant to win over a majority of deputies, but that Mistral does not want it and the EPR group has gone quiet on the subject, even though its votes will be decisive. Hervé Rony, director general of the authors’ rights society SCAM, told the paper the latest news was not good (translated from French). For publishers the stakes are concrete: today, a media group that suspects an AI model was trained on its articles runs into the opacity of developers, none of which discloses every protected work it has used. Without the law, RSF argues, the only route to compensation is one-to-one deals, and its September survey of 230 public agreements found that 71% involved Western English-language media.

The bill adds to French pressure on AI firms over content, after five French publisher groups filed an EU complaint against Google’s AI Overviews, and comes days after Arthur Mensch urged EU leaders in Dublin to move faster. What to watch on November 19 is whether the amendments are withdrawn, where the text sits on the Socialist agenda and whether deputies get as far as a vote.

← Back to AI4France