On June 9, Anthropic released Fable 5, the most capable AI it had ever put in public hands. Three days later it was gone.
Not throttled. Not rate-limited. Gone, for every customer on the planet, all at once.
Here is the timeline, because the speed is the whole story. Friday afternoon, Commerce Secretary Howard Lutnick sent Anthropic CEO Dario Amodei a letter citing national security authorities. The order: cut off access to Fable 5 and its restricted sibling Mythos 5 for any foreign national, anywhere, including Anthropic’s own foreign employees. The company received it at exactly 5:21pm Eastern. There was no appeal window, no public technical brief, no phased rollback. Just a conclusion.
Anthropic had no fast way to check the nationality of everyone hitting its servers. So to comply with an order aimed at foreign nationals, it did the only thing it could and pulled the plug on
all of its customers. Americans included. Everyone got quietly downgraded to the older Claude Opus 4.8 and most of them never knew why.
A model that millions of people opened on a Tuesday was unreachable by the weekend, on the word of one cabinet official.
What the government says happened
White House AI adviser David Sacks gave the administration’s version. Fable is the consumer release of Anthropic’s Mythos-class models, which carry serious cyber capability. A “trusted partner” testing Fable found a way to jailbreak its safety guardrails. The administration asked Amodei to either fix the jailbreak or take the model down. He declined. So they reached for export controls.
Reporting on X (unconfirmed by anyone on the record) claims the partner who flagged it was Amazon CEO Andy Jassy, and that a Friday-morning White House call preceded the order. Treat that part as rumor. The part that matters is on the record.
What Anthropic says happened
Anthropic’s public statement is blunt: the jailbreak surfaced a handful of minor, already-known software vulnerabilities, the kind that
other public models can find without any jailbreak at all. The company spent thousands of hours red-teaming Fable with the US government, the UK’s AI Safety Institute, and outside firms before launch. No one had found a universal jailbreak. It called the whole thing a misunderstanding and said it was working to restore access.
So you have a safety-obsessed lab arguing its model is not dangerous, and a government insisting it is, with the actual evidence sealed. Both can be partly right. Neither one is letting you see the file.
Why a software demo became a national-security event
For years, US export controls on AI meant one thing: chips. Keep the most advanced Nvidia hardware out of certain countries. Physical objects, customs forms, a logic everyone understood.
This was different. Nobody smuggled anything. The thing being controlled was access to a website. A text box got reclassified as a strategic weapon, and the people locked out included a French engineer in Paris and a grad student on a visa in Boston who happened to not hold a US passport.
That reclassification is the real event. The moment a model’s capability counts as a munition, the company hosting it stops being a software vendor and becomes an export-control checkpoint. Every API call is now potentially a product feature, a compliance event, and a foreign-policy decision at the same time.
The world saw the switch, and started shopping for a spare
Days later, at the G7 summit in Evian, the people who run countries said the quiet part out loud.
French President Emmanuel Macron warned that if the US “from one day to the next can turn off the switch,” it could wreck economies and the AI companies themselves. India’s Narendra Modi pushed for democratic nations to have unfettered access to top models. Cohere CEO Aidan Gomez framed it as bigger than market share: digital sovereignty is about
who controls the foundational technology that shapes economic and national security for decades.
The G7 floated a fix called the “trusted partners” framework, a way to grant approved nations and companies stable access in exchange for using the models to defend against rivals like China. The catch nobody could answer: who qualifies, who decides, and does a startup in Bangalore ever make the list.
The summit ended without a real answer. What it produced instead was a worldwide realization. The best AI on Earth is American, and the people who depend on it just learned it comes with a remote control they do not hold.
Where this actually leaves a normal person
Fable 5 will probably come back. Anthropic and the administration both say they want it resolved. This specific outage is likely temporary.
The precedent is not. The market just got told, in plain language, that frontier AI access is now a lever governments will pull. Other countries will accelerate their own models. Companies will start asking a fifth question when they pick a tool, on top of how smart it is, what it costs, how fast it runs, and how much context it holds: can this get switched off, and by whom.
For you, the lesson is smaller and more useful than the geopolitics. The convenience of living inside one AI is real. So is the risk. The people who stay productive through the next outage will be the ones who treated any single model like a rental car, not a house.
