At 5:21 p.m. Eastern on Friday, a directive from the Commerce Department landed at Anthropic ordering the company to bar foreign nationals from using Claude Fable 5 and Claude Mythos 5, its two newest models. Unable to gate access by nationality on systems already in production, Anthropic disabled both worldwide. The models had been live for three days.

Bloomberg, which confirmed the letter with a U.S. official, reads this as the first time the administration has used an export-control directive to force a leading A.I. company to withdraw a publicly deployed model from its existing customers. The document cited “national security authorities” without specifying the underlying concern.

The trigger, per Fortune, was a warning routed through Amazon. Its chief executive flagged a jailbreak technique in Fable 5 to the White House after first raising the matter internally with Anthropic. An Amazon spokeswoman called it “not uncommon for governments to seek our counsel on potential security risks.”

Anthropic’s account of what Amazon escalated is markedly less dramatic. The company describes “a narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws,” surfacing “a small number of previously known, minor vulnerabilities.” Anthropic notes that the same class of flaws exists in OpenAI’s GPT-5.5, which remains available.

Executives met with administration officials in Washington on Monday. No agreement was reached, according to Foreign Policy.

The dispute is the second in five months. In February, President Trump ordered federal agencies to stop using Anthropic’s models after the company declined Pentagon contract terms permitting use “for any lawful purpose.” In March, the Defense Department designated Anthropic a “supply chain risk,” a finding now being challenged in federal court. A pattern is emerging in which the export-control system, the procurement system, and the supply-chain-risk apparatus are each turned, in sequence, against the same private firm.

The timing is commercially awkward. Earlier in June, Anthropic confidentially filed for a public listing at a reported $965 billion valuation. A retroactive withdrawal order from Commerce is the kind of disclosure that doesn’t sit comfortably in an S-1.

“We believe this is a misunderstanding, and are working to restore access as soon as possible,” Anthropic said. The word choice is its own form of narrative management: a frontier lab worth nearly a trillion dollars describing a federal export-control action as a clerical error it expects to resolve.

Sources