The Shutdown That Stopped the AI World: The Full Story of How the US Government Just Disabled Anthropic's Most Powerful Models — And Why Anthropic Is Pushing Back

This Is Not Hypothetical. This Happened Last Night.

At 5:21 p.m. Eastern Time on Friday, June 12, 2026, Anthropic received a letter. It was sent from Howard Lutnick, the United States Secretary of Commerce. It cited national security authorities. And it contained an instruction that the company described as unprecedented in the history of commercial AI regulation: suspend all access to Fable 5 and Mythos 5 — Anthropic's most powerful and most recently launched AI models — by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.

Anthropic had no way to comply with a foreign-national-only suspension in real time. It has no mechanism to verify the citizenship status of every user accessing its platform simultaneously. The only technically feasible way to guarantee compliance with the directive as written was to disable both models for every customer, everywhere. That is what the company did. Within hours of receiving the letter, Fable 5 and Mythos 5 were offline for everyone. Claude's landing page carried a message that read: "Fable 5 is temporarily unavailable." A company that had launched its most significant models just three days earlier found itself, at the close of a Friday business day, shutting them down for the entire world.

Anthropic published its own statement simultaneously. It complied. It also explicitly and publicly disputed the directive. "This action does not adhere to those principles," the company wrote of the government's intervention, in a phrase whose restraint barely conceals the force of what it was describing. The most consequential AI regulatory event in the short history of commercial foundation models had just happened — and Anthropic was telling the world it believed it was a mistake.

What Fable 5 and Mythos 5 Are: The Models at the Centre of Everything

To understand why this directive matters as much as it does, you need to understand what Fable 5 and Mythos 5 are, where they came from, and what makes them different from every other AI model currently available to the public.

The story begins with Claude Mythos Preview — the model that Anthropic had disclosed in its product information provided within Claude's system prompt as "the most advanced frontier model" it has built. Mythos Preview was not made publicly available when it was developed. Anthropic's own communications described it as being held back from general release due to its extraordinary cybersecurity capabilities: the ability to identify vulnerabilities in code, explore methods of exploiting them, reason about multi-step cyberattack sequences, and provide assistance that in the wrong hands could be a genuinely powerful instrument for digital harm. The decision to not release Mythos Preview publicly was a deliberate act of AI safety caution by a company that has consistently argued that the capabilities of frontier models must be matched by appropriate safeguards before deployment.

Instead of a general release, Anthropic launched Project Glasswing — a controlled programme under which Mythos Preview was made available exclusively to a small number of vetted organisations, including critical infrastructure operators, cybersecurity firms and trusted research partners, who could use its advanced capabilities for defensive security work: finding vulnerabilities in their own systems before attackers found them, resolving software bugs at scale, hardening critical infrastructure against cyber intrusion. Mozilla reported using an earlier Mythos model to resolve hundreds of security vulnerabilities.

Fable 5 was the attempt to make Mythos-class capability safe for general commercial release. It was built on top of the same underlying Mythos model but wrapped in an elaborate system of safety classifiers — separate AI systems that detect incoming queries touching on cybersecurity, biology and chemistry, or model distillation — and route those queries to Claude Opus 4.8 instead of allowing Mythos's full capabilities to respond. More than 95 percent of Fable 5 sessions trigger no fallback at all; the classifier system operates transparently, notifying users when a query has been redirected. Before launch, Anthropic said it worked with the US government, the United Kingdom's AI Safety Institute, private third-party security organisations and internal teams to stress-test Fable's safeguards for thousands of hours. No tester found a universal jailbreak — a technique capable of broadly bypassing the model's protections across a wide range of tasks.

Fable 5 launched publicly on Tuesday, June 9, 2026, at $10 per million input tokens and $50 per million output tokens. Mythos 5 — the same underlying model with certain safeguards lifted — remained restricted to Project Glasswing partners at the same price. Three days later, both were offline.

The Jailbreak Claim That Triggered Everything

The government's directive was triggered by a specific event: another company claimed it had found a method to jailbreak Mythos — to bypass the safety classifiers that stood between the model's cybersecurity capabilities and unrestricted user access. According to reporting by Axios, confirmed by an administration official, that claim was made not to Anthropic through coordinated disclosure — the standard security research practice that gives an affected company time to assess and respond before external notification — but directly to the Commerce Department. The administration official confirmed that the Commerce Department decided to act after receiving this report, citing the alarm it generated about possible national security risks.

That decision — to go directly to the government rather than to Anthropic — is one of the most consequential details in this story, and one of the least discussed. The company that made the jailbreak claim has not been publicly named. The explainx.ai analysis of the incident noted that Amazon's researchers have been identified in some reporting as the source, though this has not been officially confirmed. What is confirmed is the sequence: jailbreak claim arrives at Commerce Department, not at Anthropic; Commerce Department becomes alarmed; Commerce Secretary Lutnick writes to CEO Dario Amodei; Anthropic receives the letter at 5:21 p.m.; Anthropic disables Fable 5 and Mythos 5 for the world.

Anthropic's own review of the jailbreak technique produced a sharply different assessment from the government's. The company reviewed a demonstration of the specific method and said it identified "a small number of previously known, minor vulnerabilities" — all of them relatively simple, all of them already discoverable by other publicly available models. Anthropic named OpenAI's GPT-5.5 explicitly: the same cybersecurity capabilities that the jailbreak allegedly unlocked in Fable 5 are, Anthropic says, already accessible through GPT-5.5 and comparable models that are not subject to any similar export control restriction. In Anthropic's framing, the government acted on a claim about Fable 5 that, even if accurate on its own terms, does not describe a capability that Fable 5 uniquely enables. The vulnerability is already in the public domain through other means.

Anthropic's statement described the demonstrated jailbreak as "a narrow potential jailbreak" — one that would unlock Mythos's cybersecurity capabilities in one specific instance rather than a universal bypass of its protections. No tester, across thousands of hours of pre-launch red-teaming involving the US government itself, the UK AI Safety Institute and multiple third parties, had found a universal jailbreak. The government's directive was triggered by a narrow technique. The response was total shutdown.

The Directive's Scope: Why Foreign Nationals Meant Everyone

The specific framing of the Commerce Department's directive — suspend access by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees — created a compliance problem that the company describes plainly in its public statement.

Anthropic cannot, in real time, verify the citizenship or national origin of every user accessing its API or consumer platform. A directive to exclude foreign nationals while serving American citizens requires either a prior vetting mechanism for every user — which Anthropic does not have and which would be extraordinarily complex to implement quickly — or the blunter instrument of disabling the models entirely and then working to establish an access restoration process. The company chose the latter. "The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance," its statement reads.

The inclusion of foreign national Anthropic employees in the scope of the directive adds a layer of complexity that has attracted significant commentary. Gary Marcus, a frequent and prominent critic of AI industry practices, said in a social media post that the directive made little sense given the US government's oft-stated position that America must stay ahead of China in the development of powerful AI systems. His specific argument: a directive of this breadth would likely convince Chinese-born AI researchers currently working for laboratories including Anthropic and OpenAI to return to China — effectively achieving the opposite of the national security goal it was designed to serve.

The workforce implications are genuine. Anthropic, like every major American AI laboratory, employs significant numbers of brilliant non-American citizens — researchers, engineers, and scientists from China, India, Canada, Europe and dozens of other countries who chose to build their careers at American AI companies. A directive that classifies those employees as security risks by virtue of their citizenship, and bars them from accessing the company's own most advanced models, is a statement about those employees that the people most affected by it will not miss.

Anthropic's Relationship With the Government: A Turbulent History

Friday's directive did not arrive from nowhere. It is the latest development in what NBC News accurately describes as "a tumultuous relationship between Anthropic and the federal government" — a relationship that has oscillated between productive collaboration and public confrontation throughout 2025 and 2026.

In February 2026, President Trump and Defence Secretary Pete Hegseth moved to bar Anthropic's products from federal agencies after the company sought stronger guardrails on how the Pentagon used its technology. That clash played out publicly before the two sides reached a negotiated resolution. The Trump administration had, in the days before the Fable 5 and Mythos 5 launch, reportedly tried to delay the models' public release. Anthropic declined. The export control letter followed days later.

The tension reflects a structural ambiguity in the relationship between the US government and the most capable AI companies: the government wants American AI to lead, and it wants to control what American AI can do and who can access it. Those two objectives — competitiveness and control — are not always compatible, and the Fable 5 and Mythos 5 situation is the most vivid illustration yet of what happens when they collide directly.

Anthropic's response has been to comply with the directive while publicly contesting its logic. The company said it would release more technical details within 24 hours — a pledge that signals it intends to make the case in public, with specific evidence, that the government's assessment of the jailbreak's significance is disproportionate. Fortune noted that the timing creates a specific financial problem: Anthropic had confidentially filed for a public listing earlier this month, valued at approximately $965 billion in a recent funding round. Government actions that suggest the company's most advanced models can be switched off at any moment by executive directive raise questions for potential IPO investors about the stability and predictability of the company's product access.

What the Industry, Researchers and Critics Are Saying

The reaction from the AI research and technology community has been wide, fast and largely alarmed — not at the prospect of AI safety regulation, which most researchers support in principle, but at the specific mechanism deployed here and the precedents it sets.

The core industry concern is the standard Anthropic itself articulated: "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. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers." Every major AI model has some vulnerability to adversarial prompting. Every model deployed at scale will have jailbreak techniques discovered for it. If the discovery of any such technique triggers an immediate government shutdown of the model, then no frontier model can remain deployed in a stable commercial environment. The standard being applied to Fable 5 and Mythos 5 is a standard that, applied consistently, would end the commercial deployment of powerful AI.

The coordinated disclosure question is equally significant. Standard practice in cybersecurity research is that when a vulnerability is found, the researcher discloses it to the affected organisation and gives them a defined window — typically 90 days — to assess and remediate before broader disclosure. The company that found the Fable 5 jailbreak apparently went directly to the Commerce Department instead. That decision bypassed Anthropic entirely, denying the company the opportunity to assess the jailbreak's actual severity and respond technically before government action was triggered. Whether that decision reflects competitive motivations, genuine security alarm, or simply unfamiliarity with security disclosure norms is a question that has not been publicly answered.

The Bigger Picture: AI Regulation at a Crossroads

The Fable 5 and Mythos 5 shutdown is the moment at which questions about AI regulation that have been theoretical became suddenly and unmistakably practical. Who decides when an AI model is too dangerous to operate commercially? On what evidence? Through what process? With what transparency? With what right of response for the company? These questions are not resolved by any existing statute. The Commerce Department acted under "national security authorities" — a broad and flexible legal basis that gives the executive branch considerable discretion and offers the affected company limited immediate recourse.

Anthropic's public statement speaks directly to this gap: "As we have stated publicly, we believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles." The company is not arguing that governments should have no power over AI deployment. It is arguing that that power should be exercised through a framework that is principled, predictable and technically grounded — rather than through a Friday afternoon letter that arrives without specific details, requires immediate total compliance, and rests on a jailbreak claim that Anthropic's own review characterised as narrow and already available through other public models.

That argument will now play out in a context that has no clear precedent. Anthropic is working to restore access. The government has not specified what it would need to see to permit that restoration. The models that were available to hundreds of millions of users on Thursday morning are unavailable on Friday night. And the entire AI industry is watching, because the precedent set here — whatever it ultimately resolves to — will define the relationship between the US government and frontier AI companies for years to come.

Sources

1. Anthropic Official Statement — Statement on the US government directive to suspend access to Fable 5 and Mythos 5 (June 12, 2026) — anthropic.com
2. CNBC — Anthropic disables access to Fable 5 and Mythos 5 to comply with government directive (June 12, 2026) — cnbc.com
3. CNN Business — Anthropic suspends all access to Mythos model after US government bans foreign nationals use (June 13, 2026) — cnn.com
4. Bloomberg — Anthropic Says US Orders Halt to Foreign Access for Fable 5, Mythos 5 AI Models (June 13, 2026) — bloomberg.com
5. NBC News — Anthropic suspends new AI models after government directive (June 12, 2026) — nbcnews.com
6. Al Jazeera — US orders Anthropic to disable AI models for all foreign nationals (June 13, 2026) — aljazeera.com
7. Fortune — Anthropic disables Fable and Mythos AI models after U.S. government bars it from giving foreigners access (June 13, 2026) — fortune.com
8. The Street — The White House sends a shocking message to Anthropic (June 13, 2026) — thestreet.com
9. BankInfoSecurity — US Pulls the Plug on Anthropic's Top AI Models (June 13, 2026) — bankinfosecurity.com
10. 9to5Mac — Anthropic pulls Claude Mythos 5 and Claude Fable 5 following US government directive (June 12, 2026) — 9to5mac.com
11. MarkTechPost — Anthropic Disables Claude Fable 5 and Mythos 5 After US Government Order (June 13, 2026) — marktechpost.com
12. ExplainX.ai — Why Did the US Gov Ban Fable 5? The Full Anthropic Story (June 13, 2026) — explainx.ai
13. ChainCatcher — The US government prohibits foreigners from using Fable 5, Anthropic issues a rebuttal (June 13, 2026) — chaincatcher.com
14. The Daily Star — Anthropic cuts access to Fable 5 and Mythos 5 over US 'national security' order (June 13, 2026) — thedailystar.net