Inside the fight over Claude Mythos 5
TL;DR
Anthropic received a US export control directive on Friday ordering it to suspend Mythos 5 and Fable 5 access for any foreign national, including non-US Anthropic employees. The company concluded that compliance would effectively mean disabling the products it had just spent a week promoting, then sent staff to Washington to lobby the Trump administration. The trigger appears to be a reported Fable 5 jailbreak. Anthropic says the issue is narrow, not unique, and comparable capabilities exist in rival models.
Nauti's Take
This is not a niche policy squabble, it is operational risk for every AI stack built on US models. If export control can yank product access overnight, builders need more than eval suites: real provider redundancy, data flows that understand nationality, and a Plan B that is not invented during the crisis call.
Briefingshow
This shows how quickly AI safety messaging can become a political weapon. Anthropic framed Mythos as unusually risky, and the government is now using that narrative to justify severe export controls. For enterprise buyers, model capability is no longer the only question; provider-level political risk matters too.
Sources