Anthropic gave the world its best AI, then Washington switched it off

The model was Claude Fable 5, and Anthropic shipped it on June 9 with the kind of numbers that make engineers sit up. It sat in a new tier the company calls Mythos class, above its usual top model. Stripe tested it on a 50 million line codebase and watched it do a migration in a day that would have taken a full team two months by hand. Anthropic put it on Pro, Max and Team plans free until June 22, which meant the moment it launched, plenty of Kenyan developers on those plans were holding frontier capability in their hands for nothing.

Then June 12 happened. At 5.21pm Eastern time, the US Commerce Department handed Anthropic an export control directive citing national security and ordered access suspended for any foreign national, whether inside or outside the United States. Read that line again because it is the whole story. Anthropic could not separate American users from everyone else in real time, so to comply it did the only thing the order allowed and switched Fable 5 off for every customer on the planet within hours. AWS pulled it from Bedrock. GitHub yanked it out of Copilot mid sentence. Refunds started going out for a product that had existed for three days.

This is where your framing has to be careful, because the easy word is recall and the easy word is wrong. Anthropic did not pull a faulty product. It is openly fighting the order. The government believes someone found a way to jailbreak the model, and Anthropic says the technique it was shown only surfaced a handful of minor, already known software bugs that freely available models like OpenAI GPT 5.5 turn up anyway without any trickery. Anthropic called the whole thing a misunderstanding and said it disagrees that a narrow jailbreak should be grounds for yanking a model used by hundreds of millions of people. Whether you buy that or not, this was a shutdown by the state, not a quality recall by the company.

For us, the part that should stick is that phrase about foreign nationals. You were not flagged as a security threat. You were not accused of anything. Your access vanished because you are simply not American, and the fight over whether you get the best tools is happening in a room in Washington where you have no seat and no vote. We have seen this shape before. The Mythos Preview model went out to twelve trusted organisations under Project Glasswing, names like Amazon, Google, JPMorgan, and not one of them was African. The pattern holds. The continent absorbs the risk of frontier AI early and reaches the front of the access queue late, if at all.

There is a quieter lesson here for anyone building in Nairobi or Lagos or Kigali. If your product runs on whatever model is newest and most capable, you have signed up for a dependency you cannot see and cannot control. The clever East African habit of squeezing maximum output from the sharpest available API just met its ugly edge, which is that the API can go dark overnight by order of a government that has never heard of your startup. Notice that Opus 4.8 stayed online the entire time. The boring, slightly less capable model you do not build your whole identity around is the one that was still answering calls.

Fable might come back. Anthropic clearly wants it back and says it is working on it. But the model returning does not erase what just got demonstrated, which is that the most powerful software of this decade can be repossessed from your hands with a letter you will never be allowed to read. So the question worth sitting with is simple. When the tool is this good and this fragile, are you actually building on it, or are you just renting it from someone who can take it back whenever their government decides you were never really the customer.