Anthropic proposes a US law letting Washington block a dangerous AI model, with a judge's approval

Deaths or serious injuries that would allow a block

50 people

Under the draft, the federal government could block a model at imminent risk of causing more than 50 deaths or serious injuries, or more than US$1 billion in damage, but only with a judge's approval.

What developers would have to do

• Test the most powerful models
• Be transparent with the public on results
• Submit to independent evaluation
• Keep a strong safety program

Developers of the most powerful models would have to test them, publish the results, submit them to independent evaluation and keep a strong safety program, a company spokesperson said.

Background

Congress has passed nothing on AI so far, and US policy under Donald Trump rests on executive orders and voluntary pledges. The proposal would also override some state laws, such as California's.

Charlie Bullock of the Institute for Law & AI called the bill fairly reasonable but weaker than the FRONTIER Act, a rival bill introduced in the House of Representatives this summer.