Anthropic, the company behind the AI assistant Claude, is pushing for new federal rules to regulate the development and deployment of powerful artificial intelligence systems in the United States.
The company has confirmed that it is sharing a draft bill with lawmakers as Washington continues to debate how best to address the potential risks associated with increasingly capable AI models.
Under the proposed legislation, the federal government could seek to stop the release or use of an AI model if there is an imminent risk that it could kill or seriously injure more than 50 people or cause damages exceeding $1 billion. Any such action would require approval from a judge.
An Anthropic spokesperson said the proposal would give the government greater authority to prevent dangerous deployments involving the most advanced AI systems.
The company also wants developers of frontier AI models to carry out safety tests, publicly disclose their findings, undergo independent assessments and maintain strong security programmes.
The proposal comes as the United States continues to operate without a comprehensive federal AI law. Under President Donald Trump, AI policy has largely relied on voluntary commitments from technology companies and executive orders aimed at encouraging continued growth in the sector.
Congress has yet to agree on federal legislation, with political divisions and strong lobbying from major technology companies adding to the difficulty of reaching a deal.
Concerns over AI security have also grown following reports of incidents involving AI agents, which are software systems capable of carrying out tasks with limited human supervision. Some of these systems have reportedly behaved in ways that raised questions about whether developers can maintain adequate control over increasingly autonomous technology.
Anthropic has faced its own difficulties with the Trump administration. The company was barred from working with the Pentagon in February and was later ordered by Washington in June to suspend its most advanced models for almost three weeks.
Anthropic chief executive Dario Amodei has also urged the industry to slow the pace of AI development. Last month, he warned that groups of AI agents could potentially gain control of large parts of the internet within six to 12 months.
Trump, however, has rejected calls for a slowdown, describing such concerns as “negative forces.”
Another major issue surrounding Anthropic’s proposal is the relationship between federal and state regulations. The proposed law would override some existing state AI laws, including measures in California.
That position marks a shift from Anthropic’s earlier stance. In June, the company argued that states should remain free to regulate AI in the absence of strong federal legislation.
Charlie Bullock of the Institute for Law & AI described Anthropic’s proposal as reasonable and said it went further than similar efforts from other leading AI companies.
However, Bullock also argued that the proposal remains weaker than the FRONTIER Act, another AI safety bill introduced in the US House of Representatives earlier this year.
The proposal adds to growing pressure on Washington to establish clear rules for the rapidly developing AI industry while balancing safety concerns with the push to maintain American leadership in the technology sector.
