US Export Ban Pulls Anthropic's Fable and Mythos Models Offline

The US government has forced Anthropic to take its two newest frontier models offline for every user in the world. A single export-control letter, delivered on a Friday afternoon, did what no court order had done before.

What the Directive Actually Ordered

The US Commerce Department invoked an export-control directive barring any foreign national from accessing Fable 5 and Mythos 5. Anthropic said in its statement that the order covered foreign nationals inside and outside the United States, including its own foreign-national employees.

That scope made a clean carve-out impossible. To comply, Anthropic disabled both models for all users worldwide, while leaving every other Claude model untouched.

The letter arrived at 5:21pm ET on Friday, June 12. The Hill reported that the directive cited national security concerns without spelling them out.

The Stated Reason Is a Jailbreak Anthropic Disputes

Anthropic says its understanding is that the government believed it had found a way to bypass, or "jailbreak," Fable 5. The company describes the alleged technique as asking the model to read a specific codebase and fix any software flaws.

Anthropic reviewed a demonstration and called the result minor. It said the same vulnerabilities are discoverable by other publicly available systems, including OpenAI's GPT-5.5, without any bypass at all.

According to the same CNBC reporting, the Wall Street Journal traced the intervention to Amazon CEO Andy Jassy, who reportedly told officials that Amazon researchers had circumvented some of Fable 5's anti-hacking guardrails. TechCrunch's security desk noted that the underlying letter has never been made public, leaving the real rationale unconfirmed.

A Nine-Day-Old Model, Suddenly Gone

Fable 5 was only days old when the order landed. Anthropic had launched it on June 9 as the first publicly available model in its Mythos-class line, built on the Claude Mythos Preview it limited in April under a cybersecurity program called Project Glasswing.

Any company that had wired automation onto Fable 5 lost its engine in an afternoon. The New Stack reported that a cluster of open-weight coding models became an instant fallback, because agent tooling now treats a hosted model as a swappable endpoint rather than a fixed dependency.

That replaceability is the quiet lesson for operators. When one supplier gets cut off, the teams that already second-sourced keep shipping while everyone else scrambles.

The Security Community Pushed Back Hard

The directive drew an unusually direct rebuke from the people it was meant to protect. Dozens of cybersecurity experts wrote to Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross arguing the ban took the best models away from defenders without reducing any real risk.

Their core point was about substitutability. Mythos-class models are good at finding and weaponizing flaws, the signatories wrote, but they are not uniquely good, and researchers use other foundation and open-source models for the same red-teaming work every day.

Anthropic made the same argument in its own defense. It warned that recalling a commercial model over a narrow, non-universal jailbreak would, if applied across the industry, essentially halt all new model deployments for every frontier provider.

Allies Are Treating This as a Sovereignty Warning

The most lasting damage may be diplomatic. By the G7 summit at Evian, France on June 17, the ban had become a live argument about whether allies can rely on American AI.

Canadian Prime Minister Mark Carney told reporters that the situation showed the danger of over-reliance and that allies would be wrong to accept it without diversifying. Former French minister Bruno Retailleau went further, calling the episode a "wake-up call" that a nation dependent on others for its technology can be unplugged overnight.

That framing turns a compliance story into an industrial-policy story. It mirrors the long-running debate over who actually gets to regulate AI, now playing out across borders rather than only inside Washington.

The Backdrop Makes the Timing Sharper

Anthropic is valued near one trillion dollars and filed confidentially for an IPO on June 1, so a forced product recall lands at a delicate moment. The company has also spent years asking for binding AI safety rules, and CEO Dario Amodei published an essay arguing exactly that just before the directive arrived.

The irony is not lost on the industry. A firm that asked for the power to block unsafe models became the first to have that power used against it, through a process it says was neither transparent nor grounded in technical facts.

For now, Fable 5 and Mythos 5 remain suspended. Anthropic says it believes the order rests on a misunderstanding and is working to restore access, while the Claude lineup keeps running on its other models.

The open question is whether any of this gets resolved cleanly, or whether the first government-forced model recall becomes a template. Every frontier lab is now watching how far the precedent reaches.