According to the draft, which Politico first reported, the government would be able to stop the launch or use of an AI model if more than 50 people were at risk of being killed or seriously injured, or if it caused more than $1 billion in damages.
According to the proposal, independent investigators would regularly review AI models and report any discrepancies to the US Department of Commerce.
Anthropic has previously warned of unintended consequences from AI and urged the industry to slow the pace of development.
President Trump has so far dismissed warnings from within the industry.