Anthropic’s draft federal AI safety bill draws criticism, and Dario Amodei writes a long reply to Encode’s Nathan Calvin

Politico reported on October 16 that Anthropic had pitched some congressional offices a draft federal safety bill for frontier AI models. Per Reuters and TechCrunch, evaluators registered with the Commerce Department would periodically assess labs and their models, and labs could pay them as long as payment does not depend on the results. An evaluator finding an imminent risk would have 72 hours to report it to Commerce, which could ask a federal judge in Washington, D.C., to halt the model. The risk must stem from a bioweapon, loss of control or a cyberattack, and involve more than 50 deaths or serious injuries, or over $1 billion in damage. An Anthropic spokesperson said the draft reflects the Advanced AI Framework the company has promoted since June.

Critics objected that the draft preempts some state laws and bars any other federal agency from halting a model. Per The Next Web, California State Senator Scott Wiener wrote that it “falls far short of protecting public health and safety.” On October 18, Anthony Cimino, Anthropic’s federal affairs lead, said the preemption language would be updated and the thresholds could be re-evaluated.

On October 19, Encode’s Nathan Calvin listed nine recent Anthropic actions on X, arguing that safety had become secondary; the last item was the draft bill. Dario Amodei replied at length on October 21 with nine commercially costly safety moves. They included backing SB 53, strict biosecurity classifiers and holding Mythos-class models from general release from February until June. Amodei agreed that three of Calvin’s items, including the bill, could have been handled better and said he takes responsibility.