The Trump administration asked Anthropic to take down its latest AI models, Fable and Mythos, on its own within 24 hours before imposing export controls, but was rebuffed, Politico reported on June 13 local time.

The report said that on June 11, two days after Fable was released, Amazon CEO Andy Jassy (앤디 재시) conveyed concerns to the White House about bypassing the Fable model’s guardrails. Senior White House officials, including Treasury Secretary Scott Bessent (스콧 베센트), cyber director Sean Cairncross (숀 케언크로스) and chief of staff Susie Wiles (수지 와일스), then held an emergency meeting.

The Trump administration tried to contact Anthropic CEO Dario Amodei (다리오 아모데이), and Amodei took the call after about 1 hour 15 minutes. This was followed by three calls involving about 6 senior officials, including Bessent, Cairncross and Commerce Secretary Howard Lutnick (하워드 러트닉).

During the calls, Amodei pushed back, saying the guardrail bypass at issue was limited to a specific method, not a broad “jailbreak”, and was not a serious security threat.

The Trump administration had the Amazon findings reviewed by the National Security Agency (NSA) and judged there was sufficient evidence. Bessent told Amodei directly, “You’re making the wrong decision,” but when Amodei refused, the administration immediately triggered export controls barring foreign access to the Fable 5 and Mythos 5 models.

The two sides’ accounts differ.

The White House said it chose export controls as a last resort after trying for hours to persuade the company. Anthropic countered that the White House issued an ultimatum to take down the model within 90 minutes without specific threat information. Anthropic said the measure was excessive given the severity of the issue, but it would comply with the government order.