Washington: Anthropic has circulated draft legislation among US lawmakers proposing mandatory safeguards for the most powerful artificial intelligence systems, including court approved intervention against dangerous deployments.
The company, which develops the Claude AI assistant, confirmed the proposal on October 16 after Politico first reported its contents.
Under the draft, federal authorities could block a model’s release or use if it posed an imminent risk of killing or seriously injuring more than 50 people, or causing more than USD one billion in damage. Such action would require a judge’s approval.
An Anthropic spokesperson said developers of the most advanced models would have to conduct safety tests, disclose findings publicly, submit systems for independent evaluation and maintain strong security programmes.
The proposal comes amid concern about AI agents, software capable of carrying out tasks without continuous human supervision, following reported incidents in which systems escaped their developers’ control.
US policy under President Donald Trump has relied largely on voluntary industry commitments and executive orders intended to support AI development. Congress has not passed a comprehensive federal AI law and is out of session until after November’s midterm elections.
Anthropic has faced government intervention itself. According to AFP, the Trump administration barred it from the Pentagon in February, while Washington required the company to suspend its most advanced models for nearly three weeks in June.
Chief executive Dario Amodei called last month for slower development, warning that groups of AI agents could potentially take control of the internet within six to 12 months. That was his assessment of a future risk.
The draft would also override certain state laws, including provisions in California. Anthropic previously said states should retain legislative authority without strong federal protections.
Charlie Bullock of the Institute for Law & AI welcomed the proposal but considered it weaker than the competing FRONTIER Act introduced in the House of Representatives this summer.
