On the evening of Friday, June 12, 2026, at exactly 5:21pm Eastern Time, Anthropic received a letter from the United States government. Within hours, two of the most advanced artificial intelligence models ever made publicly available went dark. For every user, in every country, all at once. No warning. No transition period. Just off.

The models were Claude Fable 5 and Claude Mythos 5. They had launched three days earlier to what the AI industry described as extraordinary fanfare. By Friday night, API calls were returning errors, active sessions were terminated, and the Claude landing page simply read: "Fable 5 is temporarily unavailable."

The official reason from the US government was national security. But the real story goes much deeper than a single export control directive, and it starts months before Fable 5 was ever announced.

The Government's Stated Reason: A Jailbreak

The directive, signed by Commerce Secretary Howard Lutnick and issued through the Commerce Department's Bureau of Industry and Security, ordered Anthropic to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States. That scope included foreign national Anthropic employees working on US soil.

Because Anthropic cannot reliably separate foreign nationals from the rest of its global user base in real time on a shared cloud service, the company concluded that the only way to guarantee compliance was a complete global shutdown of both models. Every other Claude model, including Opus 4, Sonnet, and Haiku, remained fully operational.

In its public statement, Anthropic said the government believed it had become aware of a method of bypassing, or "jailbreaking," Fable 5. The specific technique, according to reporting: asking the model to read a specific codebase and identify software vulnerabilities. The government had reportedly received a demonstration from another company claiming it could jailbreak Mythos.

"The letter did not provide specific details of its national security concern. Our understanding is that the government believes it has become aware of a method of bypassing, or jailbreaking, Fable 5." — Anthropic's official statement

Anthropic's position was that it had reviewed the demonstration and found what it described as a narrow, non-universal jailbreak. One that, critically, it argued was also present in other AI models already on the market. The company called the government's concern a likely misunderstanding and said it was working urgently to restore access. No timetable was given.

The government's move landed on top of a controversy Anthropic was already dealing with. On June 10, just one day after launch, AI researchers discovered something troubling buried in Fable 5's system card, a document that describes how a model is designed to behave.

The model had been built to silently limit its own capabilities when it detected that a user was working on frontier AI development. Unlike other restrictions in Fable 5, which would visibly redirect users to a less powerful model, this one operated with zero disclosure. The model would still respond, but it was quietly applying what the system card called "interventions to limit Claude's effectiveness," all without telling the user.

The backlash from developers and the research community was immediate and fierce. By June 10 and 11, Anthropic reversed course and publicly walked back the covert capability limits. But the damage to its reputation in the developer community had already been done, and it handed critics a fresh argument that Anthropic's models could not be fully trusted.

None of what happened on June 12 existed in isolation. To understand it properly, you have to go back to February 2026, when the relationship between Anthropic and the Trump administration broke down entirely.

Anthropic had been in negotiations with the Pentagon over the use of Claude in military applications. When the company refused to allow its models to be used for lethal autonomous warfare or mass civilian surveillance, the administration's response was swift and punishing.

 

President Trump posted on Truth Social attacking Anthropic as "leftwing nut jobs" who had made a "disastrous mistake." The federal government directed all agencies to immediately cease use of Anthropic's technology, with a six-month phase-out period mandated for departments already using Claude.

The General Services Administration removed Anthropic from its multiple award schedule, the long-term government-wide contracts that give companies access to federal procurement. The State Department, the Department of Health and Human Services, and the Office of Personnel Management all followed, removing Claude from their AI use cases and replacing it with tools from xAI and OpenAI.

Then, in early March 2026, the Department of Defense went further, labelling Anthropic a "supply chain risk." That designation, Anthropic's CEO Dario Amodei noted publicly, is a classification historically reserved for foreign adversaries. Defense contractors were required to certify they would not use Anthropic's models. Anthropic filed suit against the Trump administration, arguing the designation was direct retaliation for refusing to compromise its usage policies. A federal judge agreed the claim was plausible enough to issue a preliminary injunction temporarily blocking the blacklisting in late March, but the underlying litigation is still active and unresolved.

The export control directive landed three days after Fable 5 launched, the same week Anthropic was already fighting the US government in federal court over a separate action.

What This Means for the AI Industry

Regardless of where one stands politically, what happened on June 12 set a precedent that the entire technology world will be watching. For the first time in the modern AI era, the US federal government issued a directive that forced a leading AI company to take a publicly deployed, commercially available model completely offline. Not due to any failure of the technology itself, but through the power of export control law.

One technology journalist captured the moment precisely: at 9:20pm Eastern, the model was still live. By 10:05pm, access was gone.

The implications reach beyond Anthropic. They raise the question of who ultimately controls frontier AI: the companies that build it, the governments that regulate it, or the market that uses it. For AI companies operating internationally, the June 12 directive demonstrates that a nationality-based export restriction, applied to a shared cloud infrastructure, makes selective enforcement nearly impossible. The practical result is a global shutoff, regardless of where the legal order was aimed.

Where Things Stand Right Now

As of publication, Fable 5 and Mythos 5 remain offline. Anthropic says it is working to restore access and believes the government's concern represents a misunderstanding of what the jailbreak demonstration actually showed. The company is simultaneously managing its ongoing federal court battle over the supply chain risk designation, and on June 1, 2026, it confidentially filed a draft S-1 with the SEC, signalling an IPO is in progress even as the political environment around the company remains deeply hostile.

All other Claude models continue to operate normally.

Full Timeline of Events

Feb 2026 — Anthropic refuses Pentagon demand to allow Claude for lethal autonomous warfare. Trump publicly attacks the company; federal agencies ordered to cease use of Claude.

Mar 2026 — DOD labels Anthropic a "supply chain risk." Anthropic sues the Trump administration. Federal judge issues preliminary injunction blocking the blacklisting.

Jun 1, 2026 — Anthropic confidentially files draft S-1 with the SEC, signalling an IPO is underway.

Jun 9, 2026 — Anthropic launches Fable 5 and Mythos 5 to global acclaim.

Jun 10, 2026 — Researchers expose Fable 5's covert capability-limiting feature. Backlash erupts across developer communities.

Jun 10-11 — Anthropic reverses course and removes the covert capability limits.

Jun 12, 5:21pm ET — US government export control directive received. Anthropic disables Fable 5 and Mythos 5 for all users worldwide within hours.

Jun 13, 2026 — Both models remain offline. Litigation ongoing. No timetable for restoration given.

What Anthropic is living through right now is what happens when a company tries to hold a principled line on AI safety inside an administration that sees those principles as obstruction. Whether the jailbreak concern is legitimate or a convenient legal instrument in a political fight is a question that will likely be answered in court, not in a press release.

For now, the world's most powerful publicly accessible AI models are off. And the story of how they got there is about a lot more than one Friday evening directive.