A European client can lose access overnight to an American model it calls via API, meaning via a software interface. On June 12, 2026, a directive from the Department of Commerce required a license for any access by a foreign person to Claude Fable 5 and Mythos 5. Anthropic cut off both models for all its clients, and the reason given at the time, a circumvention of safeguards, has since been replaced by another.
According to a note from the law firm Mayer Brown dated June 30, the letter signed by Commerce Secretary Howard Lutnick imposed a license for any export of the two models, ‘including to any foreign person worldwide.’ Foreign employees of Anthropic in the United States were affected. The firm writes that the letter treats remote access via API as a ‘release’ subject to export controls. It was not published.
Previous opinions from the Bureau of Industry and Security (BIS), the U.S. export control administration, concluded that remote access to cloud software, without technology or source code transfer, is not an export. Mayer Brown calls ‘unprecedented’ a global license requirement targeting any foreign person. An Anthropic client, Legion LegalTech, whose employees work in Canada, asked a federal court to overturn the directive. The status of this proceeding could not be established.
What Emil Michael Said to The Information
Emil Michael is Under Secretary of War for Research and Engineering and Chief Technology Officer of the Department of War, the new name for the Department of Defense, according to his official biography. He is also Vice President of the Super Intelligence Force, announced on October 4, where he is in charge of relations with labs, according to The Next Web.
According to The Information, Michael commits to using executive tools, including export controls, against AI companies that do not cooperate with the government. The article, behind a paywall, could not be accessed: these remarks are known from its public summary and the account by The Next Web. Michael considers this approach appropriate and recalls past limits on cryptography exports. Existing laws are in his view ‘more than sufficient.’ ‘I think you can count on this administration, on this president, to act with urgency if there were a real problem,’ he said.
Anthropic, the Cited Example
The accounts do not name any target other than Anthropic. According to Michael’s account reported by The Next Web, the company refused to withdraw a model whose safeguards had been circumvented, then was placed under export control two hours later. Google and Elon Musk accepted, he emphasizes, the department’s terms of use.
On June 13, White House advisor David Sacks claimed that Dario Amodei had refused to fix the flaw or withdraw the model. Anthropic had responded that it had only received verbal proof of a ‘potential and narrow’ circumvention, which had been used to find ‘a small number of minor, already known vulnerabilities.’
The controls were lifted on June 30. According to Anthropic’s announcement on the same day, Mythos 5 was initially only reopened to ‘a set of U.S. organizations,’ with government approval from June 26. Fable 5 returned for everyone on July 1 on Anthropic’s services.
The dispute with the Pentagon is older. Secretary of War Pete Hegseth designated Anthropic as a risk to the department’s supply chain on March 3, 2026. The U.S. Court of Appeals for the District of Columbia Circuit rejected the company’s appeals on September 25, by a vote of two to one. According to The Next Web, Anthropic is requesting a review by the full court.
Michael does not envisage any discussion about the company’s return as long as it maintains, in its models’ constitution, a principle he disputes. ‘If you don’t want humans to lose control, you can’t give models control over humans,’ he said. He says he continues to speak to the company.
