Anthropic proposes AI law that would give a judge the final say

In the absence of federal law, Anthropic, the creator of the AI ​​assistant Claude, is proposing to American elected officials its own text to regulate AI: Washington could block a dangerous model, but only with the approval of a judge.

According to the text, revealed by the media Politico, Washington could block a model in the event of an imminent risk of causing more than 50 deaths or serious injuries, or more than a billion dollars in damage, but only with the approval of a judge. The project would also require developers to “test the most powerful models”, to be “transparent to the public” about their results, to “submit them to an independent evaluation” and to “maintain a solid security program”, indicated a spokesperson this Friday, October 16.

Under Donald Trump, American policy on AI is essentially based on decrees and voluntary commitments from companies, so as not to slow down a boom which is supporting the country's economy. Divided, Congress has not voted on anything to date on the subject, which became heated after revelations last summer of a series of security incidents caused by cutting-edge AI, some of which escaped the control of their designers. Anthropic, which was banned from the Pentagon by the Trump administration in February, was then forced in June by Washington to suspend its most advanced models for nearly three weeks.

“Our proposal also gives the federal government the ability to block or deter dangerous deployments of the most powerful models,” said the company spokesperson, whose relations with the Trump administration have recently warmed. Anthropic boss Dario Amodei called on the industry to slow down in mid-September, fearing that a group of AI agents would be able to “take over the entire internet” within six to 12 months.

Donald Trump described the multiple calls to slow down as “negative forces”. David Sacks, its former AI advisor, accuses Anthropic of using regulation to hinder its competitors. The proposed text would also take precedence over certain state laws, such as that in force in California. In June, Anthropic wrote that “in the absence of a strong federal law, states should retain full authority to legislate.”

Charlie Bullock, from the Institute for Law & AI think tank, praised X as a “rather reasonable” text, “more than what the other laboratories have done”. However, he considers it “weaker” than the FRONTIER Act, a competing bill presented this summer to the House of Representatives.