"They screwed us": Personality clashes sent Anthropic's models offline

Anthropic Suspends Mythos and Fable Models After Commerce Department Export Controls

Anthropic removed its Mythos and Fable AI models from public access after the Commerce Department issued a formal export control directive on June 12, 2026. The directive required suspension of access to Mythos 5 and Fable 5 for any foreign national. The action followed reports of potential security vulnerabilities and prior access arrangements involving entities outside the United States.

Amazon Chief Executive Andy Jassy contacted Treasury Secretary Scott Bessent on June 11 to raise concerns that the models could be modified to bypass existing safeguards. Administration officials stated that Anthropic had been informed of risks associated with certain deployment decisions and proceeded with distribution. A person familiar with the administration’s position described the company’s choices at multiple decision points as inconsistent with government expectations.

Anthropic stated that it had received explicit approval from government reviewers before deploying Fable. The company also said it had coordinated with officials on expanding Mythos access, including to a global telecommunications firm, and had revoked access in that instance prior to any formal order. A source close to the company said it did not decline to address identified issues.

The Commerce Department directive established the immediate requirement to restrict foreign access. Anthropic complied by taking the models offline. Administration officials had previously indicated possible export controls after learning that Mythos had been made available to an entity with reported ties to the Chinese Communist Party. The company maintained that all expansions of access occurred with government awareness.

Earlier disagreements between Anthropic and the Pentagon involved separate negotiations over model use. A White House official described those discussions as unrelated to the current export controls. Both episodes involved differing assessments of communication and responsiveness, according to individuals familiar with the exchanges.

Anthropic published a blog post addressing the Amazon report. It also retained a cybersecurity consultant whose prior affiliations drew criticism from administration sources. Chris Krebs, recently dismissed from a government position, publicly referenced the consultant’s involvement.

The White House had previously sought to improve relations with Anthropic after the Pentagon matter. Officials noted that rapid technical developments have outpaced regulatory processes, leading to broad measures rather than targeted adjustments. A person familiar with internal discussions said the government’s priority remained preventing unauthorized modifications while preserving U.S. technological leadership.

Scheduled meetings include a June 16 session at the Commerce Department with Anthropic technical staff members Logan Graham, Dave Orr, and Nicholas Carlini. Additional discussions are planned with the CIA and White House science adviser Michael Kratsios to address compliance with an existing cybersecurity executive order.

Administration sources said options under consideration include technical requirements to reduce the feasibility of unauthorized modifications. One official noted that complete resistance to such modifications may not be achievable and that sustained coordination between the company and regulators would be necessary. Anthropic has stated its primary objective is continued model development under applicable rules.