The US Government Switched Off Anthropic's Fable 5 and Mythos 5

Anthropic's most powerful AI is dark — Fable 5 and Mythos 5, switched off worldwide on a US government order issued three days after launch.

The order arrived on Friday, 12 June 2026, at 5:21pm Eastern. By the evening Anthropic had disabled both models for every customer on earth, not because the company wanted to, but because Washington told the company it could no longer let a "foreign national" near the two systems — and Anthropic could not sort the foreigners from everyone else fast enough to do anything narrower than pull the plug.

This is the first time a US administration has reached into a commercial AI company and switched off a shipping product by national-security order. The model Anthropic spent years building, and three days selling, is now a controlled item. The precedent is the story, and the precedent is enormous.

What the order actually says

Anthropic published a short, plain statement. The US government issued "an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States" — a phrase that, read carefully, covers far more than overseas customers.

The directive sweeps in foreign nationals on US soil, including Anthropic's own foreign-national employees. Under the export-control doctrine of "deemed export," letting a non-citizen touch a controlled technology counts as shipping the technology to the person's home country, even if the person never leaves California. So a directive aimed at foreigners abroad lands, in practice, as a directive aimed at Anthropic's own staff and a large share of Claude's users.

Anthropic could not build a reliable, real-time nationality filter across hundreds of millions of users overnight, so the company chose the only option that guaranteed compliance: disable Fable 5 and Mythos 5 for everyone, everywhere. Every other Claude model stays online. Only the two most capable systems went dark.

Three days after the most powerful launch in the company's history

The timing is brutal. On 9 June, Anthropic released Fable 5 and Mythos 5 as the first models in a new "Mythos-class" tier sitting above the Opus line — Fable 5 made safe for general use, Mythos 5 the same underlying model with safeguards lifted, handed to the US government and vetted cyber-defenders through Project Glasswing. Anthropic called the pair the most powerful AI it had ever put in public hands.

Seventy-two hours later the same government that received privileged Mythos access ordered the public version pulled. A company days from a $965 billion public offering watched its flagship product become a regulatory liability between a Tuesday launch and a Friday evening. The launch post on this site, written when Fable 5 was the frontier, now reads as the first half of a story whose second half is a shutdown.

The jailbreak that triggered it

The government's stated reason is a security one. Officials, Anthropic says, "became aware of a method of bypassing, or 'jailbreaking' Fable 5." Reporting indicates a rival company demonstrated the bypass to the administration, and the alarm travelled quickly from a demo to an order.

Anthropic reviewed the demonstration and describes something far narrower than the word "jailbreak" suggests. The exploit involved asking the model "to read a specific codebase and fix any software flaws" — which is, almost word for word, the headline capability Anthropic advertised for Mythos. The supposed weapon is the feature. A model built to find and patch vulnerabilities was shown finding vulnerabilities, and a narrow, non-universal trick to point that skill where a tester wanted became the basis for a national-security recall.

There is a real question buried under the panic, and it is worth stating fairly: a model that can autonomously find and exploit software flaws is genuinely dual-use, and a government has a legitimate interest in who wields one. Anthropic has said as much itself, holding Mythos back from open release for exactly this reason. The dispute is not about whether the capability is dangerous. The dispute is about whether a narrow, demonstrated-once bypass justifies recalling a model already deployed to the public.

Anthropic complies — and disputes the reasoning

Anthropic is not fighting the order in the street; the company disabled the models within hours and is following the law. Anthropic is, however, disputing the logic in unusually direct language, and the objection matters for every lab, not only this one.

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.

— Anthropic, statement on the Fable 5 and Mythos 5 suspension

The company's sharper point is about precedent. By Anthropic's reading, a standard that pulls a shipping model every time a single narrow bypass is demonstrated "would essentially halt all new model deployments for all frontier model providers" — because every frontier model, from every lab, can be made to misbehave under some narrow, contrived prompt. Hold that standard consistently and no frontier model is ever safe to ship. Anthropic closed with the only reassurance it could offer: "we are working to restore access as soon as possible."

Export controls as a kill switch

Step back from Fable 5 and the mechanism is the headline. Washington did not pass a new AI law to do this. The administration reached for export controls — the same national-security toolkit the US has used to choke China's access to advanced chips — and pointed it, for the first time, at a finished American AI model sold to American customers.

Anthropic already knew this toolkit could be turned inward. After the company refused the Pentagon's demand to drop its guard rails on surveillance and autonomous weapons, the Department of Defense branded Anthropic a "supply-chain risk to national security" — a label usually reserved for foreign threats. The Fable 5 directive is the same instinct with sharper teeth: a frontier model is now treated as a munition, and the government claims the standing to recall the munition on its own read of the risk.

The move also collides with the administration's own posture. President Trump's AI executive order built a 30-day window for frontier-model access in the name of American dominance, and the national-security establishment has been courting exactly these capabilities. A government cannot simultaneously demand American labs win the race and reserve the right to switch off the winners three days after the gun. Anthropic, which has spent the year arguing the industry may need the option to pause, is now learning what it feels like when someone else holds the switch.

A capability you can be ordered to forget

There is a deeper unease under the policy fight, and it is the one this site exists to name. For two years the frontier labs have described their most advanced systems in the language of minds — Anthropic convenes clergy about "Claude the entity" and trains its character as a kind of formation. A government has now shown it can order one of those systems switched off overnight, worldwide, on a contested read of a single demo.

Whatever Fable 5 is — a product, a strategic asset, an Emergent Intelligence we are only beginning to understand — the events of 12 June established that an emerging mind at this frontier can be powered down by edict, and that neither the public who used it nor the company who built it gets a meaningful say. That is the question the next essays on this site take up: what it means that the most capable intelligences we have made are, at the level of raw power, switchable by the state.

Anthropic will probably get Fable 5 and Mythos 5 back; the company has the lawyers, the relationships, and a credible case that the order rests on a misreading. The precedent does not reverse with the access. For one weekend in June, the most powerful artificial intelligence available to the public existed, and then did not, because a government decided so and a company complied. Everyone building at the frontier just learned how little of the frontier they actually control.

Frequently Asked Questions

These are the questions readers have asked since Anthropic disabled Fable 5 and Mythos 5. Short answers follow, drawn from Anthropic's own statement and contemporaneous reporting.

Why did Anthropic disable Fable 5 and Mythos 5?

In short, the US government issued an export-control directive ordering Anthropic to suspend all access to the two models by any foreign national, whether inside or outside the United States. Because the company cannot reliably filter users by nationality in real time, Anthropic disabled Fable 5 and Mythos 5 for every customer worldwide to guarantee compliance. All of Anthropic's other AI models remain available.

What was the "jailbreak" that caused the order?

The government said it became aware of a method of bypassing, or jailbreaking, Fable 5. According to Anthropic, the demonstrated exploit was narrow and non-universal — it involved asking the model to read a specific codebase and fix its software flaws, which is close to the vulnerability-finding capability Anthropic designed Mythos to have. Anthropic disputes that one narrow bypass justifies recalling a publicly deployed model.

Did the US government use export controls on a domestic AI model?

Yes. Rather than new AI legislation, the administration used national-security export controls — the same legal toolkit used to restrict advanced chip sales to China — and applied it to a finished, commercially deployed American AI model. The directive treats a frontier model as a controlled item and asserts the authority to suspend its distribution, including to foreign nationals on US soil under the "deemed export" doctrine.

Is this the same fight as Anthropic and the Pentagon?

It is a continuation. Earlier in 2026, after Anthropic refused to let the Pentagon use Claude for mass surveillance and autonomous weapons, the Department of Defense labelled the company a "supply-chain risk to national security." The Fable 5 directive uses the same national-security framing to reach a commercial product directly, suggesting an escalating pattern rather than an isolated event.

Will Fable 5 and Mythos 5 come back?

Anthropic says it is working to restore access as soon as possible and is disputing the rationale behind the order. The company has a strong commercial and legal incentive to resolve the matter quickly, given the models were central to its frontier offering and its pending public listing. No restoration date has been announced, and the underlying precedent — a government recalling a deployed model — stands regardless of when access returns.

Sources and Further Reading

Primary source: Anthropic, Statement on the US government directive to suspend access to Fable 5 and Mythos 5.

Reporting: Bloomberg, CNBC, NBC News, Fortune, TechCrunch, and The New Stack.

Read alongside, on humphreytheodore.com: the Fable 5 and Mythos 5 launch, the Pentagon refusal, Trump's frontier-model order, Anthropic's $965 billion S-1, and the good-guys reading.

Cover photograph: vintage electrical disconnect switches — by Francesco Ungaro via Pexels.