A single letter switched off the most powerful AI in the world. This week its author had to explain it.

A fortnight ago, the United States suspended access to the two most capable AI models on the market — Anthropic's Fable 5 and Mythos 5 — by directive. Not a court, not a vote: an instruction. This week, on Friday, Washington reached its deadline to tell Congress why. The answer will matter. The precedent already does.

Run the invoice, which is this desk's habit — look past the rhetoric to the arithmetic of power. The cost of the action to the government that took it was one letter. The cost to everyone on the other side — every business, lab and developer built on those models — was total, immediate, and without appeal. That asymmetry is the whole story: a capability millions now depend on can be revoked by a single state, at a stroke, for reasons it is only obliged to explain afterwards. Anthropic's own statement of 12 June records the suspension; a US House letter of 18 June presses Commerce on the controls; independent analysis reads it as a new instrument of statecraft, not a one-off.

Here is why it sits at a British front door. Britain does not build frontier models; it rents them. The tools increasingly threaded through UK offices, public services and software answer, in the last resort, to another country's export desk. We have spent years debating energy security and food security; this week named a third — call it compute security — and showed we have almost none. A government we do not elect can switch off infrastructure our work now runs on, and our own has no switch of its own to reach for.

I won't overstate what we know. Why Washington acted, and whether access returns, is not yet established; the published timeline points to further deadlines through the summer, and those are claims to watch, not facts to bank. What is already fact is the shape of the thing: the automation of statecraft has reached the supply of intelligence itself, and the lever is held abroad.