Amodei says Anthropic held Mythos back for four months for safety

Anthropic CEO Dario Amodei defended the company's "race to the top" AI safety strategy and listed nine decisions that carried commercial costs, according to The Next Web. His response followed criticism from Encode AI executive Nathan Calvin over Anthropic's handling of its draft AI safety bill.

Amodei said Anthropic delayed broad access to its Mythos class models from February until June while it strengthened cyber safeguards. He also cited strict biological safeguards, support for California's SB 53 safety law, chip export controls for China, restrictions on sales to companies controlled by China, and a public benefit fund that will hold 5% of Anthropic's shares.

Amodei said Anthropic could have handled three of Calvin's points better, blaming "rhetoric or policy communication that wasn't carefully written," and said that as CEO he takes responsibility. He called changes to the company's Responsible Scaling Policy the most substantive issue, said its previous version was too rigid, and said he would look further into running a separate compliance framework alongside it.

He also said Anthropic is nearing the point where AI moves too fast for safety work to keep up, which is shaping its thinking on "pacing," and that he would have more to say soon.