Darcos law buried, Colorado law gutted: the day AI lobbying won everything
On the same day, in Paris and Denver, two AI regulations went down. The lobbying argument was almost identical, word for word.
On May 12, 2026, two pieces of legislation died on the same day. In Paris, the Darcos law was kept off the National Assembly's agenda. In Colorado, the vote was harsher: 57 to 6, to gut the first U.S. law on algorithmic transparency. Five thousand miles apart, the same move, almost the same argument. Few people connected the dots.
And yet, both defeats tell the same story. When the AI industry decides a regulation is coming too early, it now has the political muscle to push it back. Including when that industry markets itself as the alternative to the GAFAM.
The Darcos law, in plain terms
Adopted unanimously by the French Senate on April 8, sponsored by Senator Laure Darcos with cross-party backing, the proposal had a simple goal: introduce a presumption of exploitation of cultural content by AI providers. In other words, flip the burden of proof.
Today, an author who suspects their book was used to train ChatGPT has to prove it. Good luck: training corpora are opaque, and scraping rarely leaves accessible fingerprints. Under the Darcos law, AI providers would have had to demonstrate they had not used a given work. A corrected asymmetry, not a witch hunt.
On March 19, France's Conseil d'État had validated the mechanism: constitutional, compatible with EU law. No exploitable legal weakness. All that remained was Assembly scheduling.
The Mistral argument, or how a national champion ends up defending the industry
That is when Mistral entered the picture. According to Livres Hebdo and Décideurs Juridiques, Arthur Mensch (CEO of Mistral) and Yann Le Cun (Meta) met group and committee chairs one by one. The message: the French law would isolate France within the European Union and slow domestic innovation. The technical argument: creating a presumption would open a "litigation premium" that would hit French players harder than the GAFAM.
It is a clever argument because it is exactly the one the GAFAM have been making everywhere for a decade. "Premature regulation kills innovation," "you will shoot yourself in the foot," "see what other countries do before moving." When Google or Meta say it, people get suspicious. When a French champion says it, people listen.
Laure Darcos eventually used the line that has been circulating since: in her words, Mistral acts as "a Trojan horse for the American giants." Her reading: by defending a modest lump-sum contribution rather than full recognition of rights, Mistral was objectively serving the interests of Meta, Google and OpenAI, which scrape European content at scale.
You can share that reading or find it excessive. The factual element holds: on copyright, the national champion's lobbying converged point by point with that of the GAFAM. And it won.
Colorado's mirror, the same day
The timing is almost too perfect. On May 12, while Paris is not debating, Denver is voting.
The target was SB 205, adopted in 2024. It was the first comprehensive U.S. law on algorithmic transparency. It required companies using AI for high-stakes decisions (hiring, credit, housing, healthcare) to document how their systems work, manage risk, and run impact assessments. It was set to take effect on February 1, 2026.
SB 189, passed 57-6 in the House and 34-1 in the Senate, replaces all of that with a simple notification obligation: tell the consumer an AI was used, and give them a right to appeal. No more duty of care. No more audits. No more model transparency. Effective date pushed to January 2027.
The tech lobby's arguments in Colorado? "Chilling effect on investment," "unmanageable complexity for startups," "risk of state-level isolation from the competition." The script is familiar.
What this says about the word "sovereignty"
The official French narrative places Mistral on the side of the good guys: a European champion valued at 13.8 billion euros (after ASML joined the cap table), a French team, a continental answer to OpenAI. All true. And all insufficient to make sense of what happened on May 12.
A private company worth 13.8 billion defends its corporate interests. That is its function, and nobody should be surprised. What does raise a question is the habit of talking about Mistral as if it were a public service because it is French. That slippage lets entirely standard lobbying pass for an act of patriotism.
And the confusion has a cost. On that same May 12, Arthur Mensch was testifying before the Assembly's commission of inquiry on digital vulnerabilities. Asked about AI's social impact, he acknowledged that rising unemployment in certain sectors was "not excluded," and that value was shifting "from labor to capital." The CEO of the national champion concedes the social cost while his company helps bury the law that would have protected some of those affected.
The real fault line
What the two May 12 stories show is that the meaningful question is no longer "Europe versus the United States." It is "industry versus regulation." Along that fault line, national borders matter little: a European champion and a Californian giant make the same case when a text threatens their trajectory.
For French creators who believed European technological sovereignty would shield them from OpenAI's appetite, the takeaway is harsh. The law that would have flipped the power balance is sitting in a parliamentary drawer. And the company that helped bury it speaks French.
The Assembly's next window opens in September. By then, Mistral will have raised more capital, OpenAI will have scraped more corpora, and the "litigation premium" that worried opponents so deeply will have been kicked further down the road. Like algorithmic transparency in Colorado: January 2027, or perhaps never.
