The US government has ordered Anthropic to suspend access to Fable 5 and Mythos 5, and the company says the only way to comply is to switch its most capable models off for everyone.
In a statement published the evening of June 12, Anthropic said the export-control directive arrived at 5:21 p.m. Eastern that day, citing national security authorities. It bars any foreign national from accessing the two models, whether inside or outside the United States, including Anthropic's own foreign-national employees. The net effect, the company says, is that it "must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance." Other Anthropic models are unaffected.
The order lands three days after Anthropic put Fable 5 in front of the general public for the first time, a launch we covered on June 9. The model had been included free on paid Claude plans through June 22; Mythos 5, the less-restricted variant, was limited to vetted cyberdefense partners. Both are now dark.
Why it matters: by Anthropic's account, the US government has pulled a deployed frontier model out of the hands of hundreds of millions of users by direct order, with no published directive and no public explanation. It is a preview of the fight over who gets to switch off an AI system, on what evidence, and through what process. The sharpest twist: Anthropic has publicly campaigned for the government to have exactly this kind of authority.
What the government is worried about, as far as we know: the letter itself gave no specifics, according to Anthropic. The company's understanding is that the government believes it has found a method of jailbreaking Fable 5, and that the only evidence provided so far is verbal: a narrow bypass that "essentially consists of asking the model to read a specific codebase and fix any software flaws." Anthropic says the technique surfaced "a small number of previously known, minor vulnerabilities," that rival models including OpenAI's GPT-5.5 can find the same flaws without any jailbreak, and that defenders use the capability every day. It has promised more detail within 24 hours.
The company asked for this power, just not like this: in its Advanced AI Framework, Anthropic argues that governments should have the legal authority to block or deter dangerous AI deployments, with civil penalties tied to global revenue. CEO Dario Amodei was making that case in public as recently as June 10, two days before the directive arrived, announcing an essay on the framework and writing that AI "is progressing extremely fast—much faster than the policy process was built to handle." The company is not retreating from that position. It says the authority should run through "a statutory process that is transparent, fair, clear, and grounded in technical facts," and that "this action does not adhere to those principles."
What Anthropic says it will do: comply now, argue later. The company is disabling both models, disagrees that what it calls "a narrow potential jailbreak" justifies recalling a commercial model deployed to hundreds of millions of people, and warns that applying that standard industry-wide "would essentially halt all new model deployments for all frontier model providers." It calls the episode a misunderstanding and says it is working to restore access as soon as possible.
What it looks like in the product: Anthropic's developer account posted the operational details on X Friday evening. New sessions across Claude products will run on the user's default model or Opus 4.8. Existing Fable 5 sessions "will end with an error." API requests to Fable 5 will also return errors, and Anthropic is telling developers to "update your integrations to other Claude models."
What we tested ourselves: the shutoff was not complete as of this writing. At 9:01 p.m. ET Friday, roughly an hour after the service notice, a Still in the Loop test request to claude-fable-5 on the Claude API returned a normal completion rather than an error, and a Claude Code session running on Fable 5 remained active. That is one US-based account, and it proves only that the cutoff had not finished rolling out. It says nothing about how broadly access remains.
What we can confirm from the source: nearly everything in this story comes from Anthropic: its statement and the service notice from its developer account. The directive has not been published, the government had made no public comment as of this writing, and none of the company's account is independently verified.
What we still don't know: which agency issued the directive, and under precisely what authority. What the government believes the jailbreak enables, and who demonstrated it. Whether the flaw is as minor as Anthropic says, since no one outside the company and the government has examined it. And how long it lasts.
The larger story is one Anthropic helped write itself. The company asked for a world where governments can stop a dangerous model, and it is now the test case for how that power gets used.