US orders Anthropic to pull Fable 5

On the evening of 12 June 2026 — at 5:21pm Eastern, by Anthropic’s own account — the company received a directive from the US government ordering it to suspend all access to its two newest and most capable models, Fable 5 and Mythos 5. The order cited national-security export-control authorities and barred access by any foreign national, whether inside or outside the United States. The models had launched just three days earlier, on 9 June. By the morning of the 13th they were dark — for everyone.

3 daysbetween the launch of Fable 5 and Mythos 5 and the government order that took them offline worldwide.

What the directive does

The directive, issued under national-security export controls, prohibits Anthropic from serving Fable 5 or Mythos 5 to any foreign national — including the company’s own foreign-national employees. Bloomberg reported the order came via the Commerce Department’s Bureau of Industry and Security, associated with Commerce Secretary Howard Lutnick.

Crucially, the practical effect is far broader than “foreign nationals”. A cloud AI service does not check a passport on every request, and Anthropic cannot reliably establish a user’s citizenship in real time. Faced with a rule it could not enforce selectively, the company chose the only option that guaranteed compliance: a hard shutoff for all. “We must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance,” its statement reads. Every other Anthropic model remains available.

For readers here, that is the headline: every UK developer, business and individual is a foreign national under this directive. Two of the most capable models on the market vanished overnight, with no UK input into a US decision.

The government’s reason — and Anthropic’s pushback

The stated trigger was security, not trade. Per Fortune, officials acted “after learning of a technique to bypass Fable 5’s safeguards” — safeguards “designed to prevent users from accessing the powerful cybersecurity abilities of Mythos.”

Anthropic’s account is markedly less alarming. It describes what the government demonstrated as a “narrow, non-universal jailbreak” — one that would unlock Mythos’s cyber capabilities in a single specific instance, not a universal break of all of Fable 5’s defences. It adds that the letter “did not provide specific details” of the national-security concern, and that the capability in question “is widely available from other models” and is “used every day by the defenders who keep systems safe.”

The company is unambiguous that it objects:

“We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.”

It warns the standard, applied consistently, would “essentially halt all new model deployments”. But it is complying: “We are complying with the government’s legal directive and are removing access to Fable 5 and Mythos 5 for all users.”

How an export control reaches a cloud model

Is it an overreaction?

The reaction from independent voices has been sceptical. Simon Willison, who tracks model releases closely, summarised the episode in a single word on his blog: “nuts”. Per Fortune’s reporting, policy writer Dean Ball called the move “simply cartoonish,” reading it as either narrow legal targeting or “extreme national-security hawkery”; AI critic Gary Marcus said it makes “little sense” while the US is competing with China on exactly this technology; and cybersecurity researcher Peter Girnus offered the sharpest line: “If you describe your product as a munition in every press release, eventually a government takes you at your word.”

That last point cuts both ways. Frontier labs have spent two years framing their models as potentially dangerous to justify safety investment and regulation. A government acting on that framing — recalling a model over a cyber-capability concern — is the logical, if uncomfortable, conclusion.

What it signals — and what to watch

Whether any of this makes the world safer is genuinely unclear. The directive’s logic is to deny a dangerous capability to adversaries. But by Anthropic’s own account the capability is already widely available in rival models; the trigger was a narrow, unproven jailbreak with no detailed evidence supplied; and because nationality cannot be checked per request, the result is not “foreigners lose access” but “a US company’s two best models leave the global market.” Equivalent tools remain a click away. The safety gain is, at best, hard to locate.

What is unambiguous is the precedent. For the first time, a US frontier model has been pulled from worldwide service by government order, days after launch, on national-security grounds. That reframes a question every buyer outside the US should now ask: how much of your stack depends on a single model, from a single vendor, in a single jurisdiction that can switch it off without consulting you?

Three things to watch over the coming weeks:

• Whether the order holds — or is narrowed or rescinded once Anthropic’s “narrow jailbreak” argument is tested.
• Whether it becomes a regime — whether other labs’ frontier models attract similar directives, turning a one-off into a pattern.
• Whether buyers diversify — whether UK and EU customers start keeping a capable open-weight or non-US model in reserve, so a decision made in Washington can’t take their tools offline overnight.