Anthropic proposes a law that would allow America to stop dangerous artificial intelligence models

Listen to the article. The audio text is automated, generated by an automated system

Anthropic, the developer of the artificial intelligence-based assistant “Cloud,” confirmed on Friday that it is circulating a draft law among American lawmakers aimed at regulating artificial intelligence, amid the absence of federal legislation addressing the risks posed by this new technology.

According to the draft revealed by Politico for the first time on Friday, Washington can prevent the launch or use of any artificial intelligence model in the event that there is an imminent risk of death or serious injury to more than 50 people, or causing damage exceeding a billion dollars, provided that judicial approval is obtained.

The director of the IMF warns that the risks of artificial intelligence outweigh its benefits without controls

A company spokesperson told AFP that the proposal "gives the federal government the ability to prevent or deter dangerous deployments of the most powerful artificial intelligence models."

He added that the proposal "obligates leading developers in this field to test their most powerful models, be transparent to the public about the results of these tests, subject the models to independent evaluations and maintain a strong security program."

Artificial intelligence policies in the United States currently rely mainly on voluntary pledges by companies operating in the sector, in addition to executive orders aimed at supporting the prosperity of artificial intelligence, which technology leaders see as a driver of the American economy.

As for Congress, which is politically divided and faces intense pressure from major technology companies, it has not yet passed a federal law regulating artificial intelligence, and it is currently in recess until after the midterm elections scheduled for next month.

The issue of security has become the focus of great attention following the recent revelation of security incidents caused by the behavior of “artificial intelligence agents,” which are software designed based on artificial intelligence models and capable of carrying out a series of successive tasks autonomously.

Anthropic, which was banned by the Donald Trump administration from dealing with the US Department of Defense in February, was forced to suspend the work of its most advanced models for about three weeks, under pressure from Washington.

Last month, Anthropic CEO Dario Amodei called on the industry to slow down the pace of development, fearing that a group of artificial intelligence agents could take over “the entire Internet” within 6 to 12 months.

In contrast, Trump described repeated calls to slow the development of this technology as “negative forces.”

The proposed legislation also provides for overriding some laws at the state level, such as those currently in effect in California.

Anthropic reported in June that “in the absence of strong federal law, states should retain full authority to enact their own legislation.”

Charlie Pollock of the Institute for Law and Artificial Intelligence think tank praised the proposal as “very plausible” and “beyond what other leading AI labs have presented.”

But he said in a post on X that he sees the proposal as “weaker” than the Frontier Act, a competing bill introduced in the House of Representatives this summer.