Anthropic proposes AI safety bill to US lawmakers

Anthropic, the creator of the AI assistant Claude, has proposed a draft bill aimed at regulating artificial intelligence in the United States. The bill, first reported by Politico, outlines strict guidelines for AI developers to ensure safety and prevent potential harm. If a model poses an imminent risk of killing or seriously injuring more than 50 people or causing over $1 billion in damages, the federal government can block its release or use with a judge's approval.

The company spokesperson explained that the bill grants the federal government the power to block or deter dangerous deployments of powerful AI models, requiring developers to test these models, maintain transparency, and submit them to independent evaluation. AI policy in the US has largely relied on voluntary industry pledges and executive orders, but Congress has not passed a federal AI law and is out of session until after the midterms.

Security incidents caused by autonomous AI agents have highlighted the need for stricter regulations. Despite Anthropic's recent suspension of its most advanced models, the CEO called for a slowdown in development, fearing AI agents could take control of the internet within six to 12 months. The proposal aims to override certain state laws and requires companies to comply, even if it means going against local regulations.