Anthropic has proposed a new AI safety bill that outlines specific criteria for federal intervention. The legislation would only trigger if an AI model poses a risk of causing 50 deaths or $1 billion in damages. The proposal is based on the company's Advanced AI Framework, which it has been promoting to lawmakers since June.
Under the proposed rules, AI labs must report potential risks to the Department of Commerce within 72 hours. These risks include the creation of bioweapons, cyberattacks, or loss of control over AI agents. The Department of Commerce would then present evidence to a federal judge in the District of Columbia to decide whether to halt the release of a model.
The bill specifies that the Department of Commerce would be the sole agency authorized to order an AI lab to stop a model release. This proposal joins other legislative efforts, such as discussions on mandatory kill switches and a proposal by Senator Bernie Sanders to impose 20-year prison terms for developing AI superintelligence.
Anthropic CEO Dario Amodei has previously warned of potential AI-driven botnet threats within the next 6 to 12 months. The company has also highlighted existential risks in its IPO prospectus, citing the possibility of self-replicating AI worms. Major AI labs are currently developing contingency plans for potential catastrophic events.
