When an AI model goes dark on command: the Fable 5 affair, seen from Europe
A US directive switched off two AI models for the entire planet. Yet the order targeted only foreign nationals: behind the cutoff lies a geopolitical stake — a strategic capability a state can ration by nationality, and the competitive imbalance it would open between the United States and the rest of the world.
Picture the scene. An ordinary Tuesday evening, 5:21 p.m. For months, your teams have woven a whole chunk of their work around one tool: customer support, document analysis, that little automation pipeline nobody can do without anymore. And then the tool goes dark. Not an outage, not maintenance: a decision, taken thousands of kilometres away, by a government that isn’t yours.
That’s roughly what played out that day. On a directive from the US government — in the name of national security and export controls — Anthropic had to suspend access to two of its models, Fable 5 and Mythos 5. And here’s a detail that’s anything but minor: the directive doesn’t target the whole world, it targets “any foreign national,” inside or outside the United States — right down to the company’s own foreign-national employees. Unable, or unwilling, to sort its users by passport, Anthropic ended up cutting access for everyone, Americans included. Visible result: a blanket cutoff. The text’s intent: to deprive foreigners.
A blanket cut, a precise target
That nuance changes everything. The cut-for-everyone is the spectacular part — and, at bottom, almost accidental: a side effect of compliance. The deliberate part, the one worth pausing on, is the principle it establishes: access to a frontier capability can now be granted or denied by nationality, on a government’s signature. This time the move backfired on its author — Americans lost the tool too. But nothing rules out a more surgical version tomorrow: the capability kept at home, denied abroad. The precedent is set; the text already points the way.
The most strategic input of the era
Because these models aren’t just another gadget. They’re becoming a basic building block, on a par with electricity or compute: an input you use to produce almost everything else. And if a single state can decide who, by their flag, gets the good version of that input, then it holds a lever over the competitiveness of entire economies.
The stake isn’t one company’s invoice; it’s whether the productivity engine of the next two decades will be available to you on the same terms as to your competitor across the water. At a firm’s level, dependence shows up in contracts and margins. Stacked up, at the scale of a continent, it becomes a strategic asymmetry — a competitive imbalance that an administrative decision, taken elsewhere, can widen overnight. A state holding such a lead has every interest in keeping it; that’s fair game. All the more reason not to bet your future on the wager that it never will.
Keeping the choice
So what do we do? Not retreat — that would mean running with a ball and chain. Autonomy, here, doesn’t mean autarky; it means keeping the choice, and the choice is prepared the way you hedge a risk, long before the storm.
In practice: diversify suppliers and slip an abstraction layer between your code and any given model, so you can swap it out without rewriting everything. Favour, where possible, open-weight models you host yourself — no one can ration those from the far shore. Know how to run things locally, as an individual and as a country. And fund a credible European alternative before you desperately need it, because an alternative can’t be improvised on the day of the cutoff.
Fable 5 will have been only a rehearsal — a misfire, this time, more than a demonstration. But it showed where the switch is, and in whose hands. The rest is a matter of clear sight: keep betting everything on a plug you don’t hold, or keep, while you still can, enough not to depend on it.
Based on the statement published by Anthropic on 12 June 2026 (anthropic.com). Model names and technical details are those of the statement.