Three days. That’s how long Claude Fable 5 was the most capable AI model the public could touch.
Anthropic shipped it June 9, live on the API and Amazon Bedrock. Friday at 5:21pm ET, Commerce Secretary Howard Lutnick sent Dario Amodei an export control letter. By that night, Fable 5 and its sibling Mythos 5 were dark. For everyone.
The order targeted foreign nationals — anyone outside the US, plus non-citizens inside it, including Anthropic’s own staff. You can’t filter users by nationality in real time, so compliance meant pulling the models outright. Opus 4.8 and the rest stayed up.
The trigger was a claimed jailbreak. Anthropic says it’s narrow — point the model at a codebase, ask it to fix the flaws — and that GPT-5.5 does the same with no export controls attached. AI policy researcher Dean Ball called the directive “cartoonish.”
What sticks with me, as someone who makes build-vs-buy calls #
This looks like the first time a leading lab has taken a live model offline on government orders. Not a deprecation. Not an outage. A model you shipped into production on Tuesday, gone by Friday, for a reason no architecture review would have caught.
The lesson isn’t “avoid frontier models.” It’s that availability is now a regulatory variable, not just an uptime number. So keep an abstraction layer and a real fallback in place, and don’t bet a critical path on one model you don’t control.
Rethinking your dependencies yet?