AI · Field Notes
I Installed Anthropic's Best Model. Three Days Later the Government Switched It Off.
Claude Fable 5 lasted 76 hours. Here's what actually happened - and why it should rattle anyone who builds on tools they don't own.

I'd barely finished wiring Claude Fable 5 into my workflow when it vanished.
Anthropic released it on June 9, calling it the most capable model they had ever made available to the public. I did what I always do with a new model: pointed a couple of real projects at it, got a feel for where it was sharp and where it wasn't, started to lean on it a little. A normal week.
By Friday it was gone. Not deprecated, not rate-limited - gone. Open Claude and it just said the model was currently unavailable.
Here's the part that surprised me. It wasn't pulled because it was broken, or unsafe, or unpopular. It was pulled because the US government told Anthropic to pull it.
What happened, in plain terms: Washington issued something called an export-control directive. Export controls are the rules that decide which technologies can leave the country or reach foreign nationals - the same machinery that governs advanced chips and, years ago, encryption. This time the "technology" was an AI model. The order said no foreign national, inside or outside the US, could touch Fable 5 or its bigger sibling, Mythos 5.
And here's the kicker: Anthropic couldn't comply halfway. There was no clean way to fence off only the foreign users, so the only way to obey the order was to switch the models off for everyone. Including me. Including every paying US customer. A rule aimed at people in other countries took the tool out of my hands in mine.
A rule aimed at people in other countries took the tool out of my hands in mine.
I want to be careful not to overclaim the why. The national-security reasoning isn't public. But what happened is clear, and it's new: a lab shipped a product, and the government reached in and turned it off three days later. Not a recall the company chose. A switch someone else flipped.
For most of computing history, "the vendor can change the deal" was the risk you planned around - price hikes, deprecations, an API that breaks your integration on a Tuesday. This is a different risk, one floor up. The vendor can be compelled, by someone neither of you elected to the decision, and your access can disappear overnight no matter what either of you wanted.
I got three days with the best model Anthropic had ever shipped. Then I got a lesson about who really controls the tools we build on. Honestly, the second one might be worth more.

