Anthropic has proposed a congressional framework that would sharply limit when federal authorities could stop an AI system, according to Truthout, citing reporting by Politico. The article describes an imminent-risk threshold involving at least 50 potential deaths or serious injuries, or financial harm reaching $1 billion. This concerns anticipated harm, not a requirement that casualties have already occurred.
According to the report, qualifying threats would also need to involve biological weapons, cyberattacks or a system moving beyond developer control. Commerce Department-supervised evaluators would assess those risks, and only that department could seek judicial intervention. Truthout identifies federal judges in Washington, D.C., as the decision-makers and raises a potential conflict if AI companies fund evaluators’ salaries.
The article quotes Anthropic defending the proposal as consistent with its Advanced AI Framework and saying it had promoted those policies to lawmakers since June. It also presents criticism calling for stronger safeguards and changes to companies’ incentives. These provisions are reported descriptions of a proposal, rather than enacted requirements or independently verified findings.