Anthropic Proposes New AI Regulation Bill to US Lawmakers

Anthropic, the company behind the AI assistant Claude, has proposed a new bill to US lawmakers to regulate artificial intelligence. This move comes as there is currently no federal law addressing the risks associated with AI technology.

The draft bill, first reported by Politico, suggests that the government could prevent the release or use of an AI model if it poses a serious risk of harming over 50 people or causing damages exceeding $1 billion. However, this would require approval from a judge.

A spokesperson for Anthropic stated that the proposal would give the federal government the power to stop dangerous uses of powerful AI models. It would also mandate that developers test these models, share their findings with the public, undergo independent evaluations, and maintain strong security measures.

Currently, US AI policy under former President Donald Trump has mostly relied on voluntary commitments from tech companies and executive orders. Congress has not yet passed a federal law on AI and is currently out of session until after the upcoming midterm elections.

Concerns about AI safety have grown following incidents where autonomous software operated outside of human control. Anthropic faced restrictions in the past, including a ban from the Pentagon and a temporary suspension of its advanced models.

Anthropic’s CEO, Dario Amodei, has previously urged the industry to slow down AI development, warning that AI agents could soon gain control over significant parts of the internet. In contrast, Trump has dismissed calls for caution as “negative forces.”

The proposed legislation would also take precedence over some state laws, including those in California. Charlie Bullock from the Institute for Law & AI commented that while the proposal is reasonable, he believes it is weaker than another bill, the FRONTIER Act, introduced in the House this summer.