🤚 The Open-Palm Recall Notice

At 5:21 PM ET on Thursday, June 12, the United States government issued an export control directive ordering Anthropic to immediately suspend all access to its two most capable AI models — Claude Fable 5 and Claude Mythos 5 — for any foreign national, whether inside or outside the United States, including Anthropic’s own foreign national employees.

Because Anthropic cannot, in real time, sort its hundreds of millions of users into “American” and “not American” bins with the confidence the federal government apparently expects, the company did the only thing it could: it pulled the plug on everyone.

By 6:59 PM Pacific Time, both models were dark. Other Claude models — Opus, Sonnet, Haiku — remain unaffected, which is the AI equivalent of telling you the restaurant is still open but the chef has been deported.

👐 The Two-Handed Jailbreak Panic

The government’s stated reason? Officials believe they became aware of a method of jailbreaking Fable 5 — specifically, a technique involving asking the model to read a codebase and identify software vulnerabilities. Which, to be clear, is something security researchers do every single day using tools that are freely available, open-source, and not subject to export control.

Anthropic’s response was the corporate equivalent of a slow blink. The company stated that the demonstration identified “a small number of previously known, minor vulnerabilities” and that “the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5), and is used every day by the defenders who keep systems safe.”

In other words: the government watched a model do what every other model already does, panicked, and grounded the most expensive one.

To make matters more procedurally delightful, Anthropic says the government provided only “verbal evidence” of the jailbreak — no written documentation, no technical analysis, no reproducible demonstration. Just a phone call and a legal order. The directive was issued under national security authority, which in practice means there is no appeals process that operates at the speed of relevance.

🌿 The Gentle Awakening

There’s a philosophical trapdoor opening here, and the entire AI industry is standing on it.

If a government can suspend a commercially deployed model — one that is integrated into enterprise workflows, API pipelines, and the daily operations of hundreds of millions of users — because someone found a jailbreak, then every frontier model is one clever prompt away from being recalled like a defective Toyota.

Anthropic made this point explicitly: applying this standard across the industry “would essentially halt all new model deployments for all frontier model providers.” Which is either a terrifying precedent or a regulatory strategy, depending on how cynical your Thursday evening has been.

The irony is exquisite. Anthropic — the company that has spent more calories on AI safety than any other lab on Earth, the company that literally invented Constitutional AI, the company that we have written about filing a trillion-dollar IPO while simultaneously asking everyone to slow down — just got told by the US government that its model is too dangerous to exist on a Thursday.

👑 The Gold-Leaf Precedent

Here is what this actually means for the industry:

• Export controls now apply to model access, not just hardware. We’ve graduated from chip bans to capability bans, and the line between “AI service” and “controlled technology” just evaporated.
• Anthropic’s compliance was total and immediate. No staged rollback, no geographic filtering, no grace period. The entire user base was cut off within two hours of the directive. This is what compliance looks like when the alternative is a national security violation.
• The “jailbreak” rationale is a loaded weapon. Every frontier model has jailbreaks. Every single one. If discovering a jailbreak is sufficient legal basis to suspend a model, then every lab in the world now operates at the pleasure of whichever government official picks up the phone next.
• Foreign nationals at US companies are now a compliance liability. The directive explicitly applies to Anthropic’s own employees. Your H-1B engineer who was fine-tuning Fable 5 yesterday is now a person who cannot legally touch the model they helped build.

Anthropic has promised to share more technical details within 24 hours. The company is working to restore access. The government has not indicated a timeline. The models remain dark.

And somewhere in Washington, someone is very proud of themselves for grounding the AI that finds bugs in code, while GPT-5.5 continues to do the exact same thing, uninterrupted, for $200 a month.

“The government found a jailbreak in our most capable model and suspended it immediately. The jailbreak was: using the model.” — The Slap of Wisdom National Security Desk, drafting this article on a model that is, for now, still legal