Anthropic Furious To Learn Its Civilization-Ending Warnings Were Taken Seriously

SAN FRANCISCO — Anthropic expressed outrage Friday after federal regulators responded to two months of the company’s own warnings about catastrophic AI risks of its new Mythos model by treating those warnings as credible and subsequently forced the AI giant to take the model offline.

The Commerce Department’s directive arrived by letter at 5:21 p.m., roughly one day after a widely shared jailbreak bypassed the new model’s much-praised safeguards and posted instructions for cyber exploits and methamphetamine synthesis. Anthropic had described the same safeguards earlier that week as substantially more effective than any model’s, ever.

In a statement, the company objected that “a narrow potential jailbreak” should not cause the recall of a model deployed to hundreds of millions of people. This is the first known instance of Anthropic arguing one of its products was not dangerous enough to warrant concern.

The company warned that applying this standard across the industry would “essentially halt all new model deployments for all frontier model providers”, an objection that the safety bar, if enforced, would inconveniently apply to everyone.

The model, marketed since April as a weapon too powerful for public hands, was removed from public hands. Anthropic said it is working to restore access as soon as possible, so that the technology it called an existential threat can resume helping people draft emails.