Two weeks ago, Washington switched off the most powerful AI Britain could buy. We have no way to switch it back on.

On 12 June, a letter reached Anthropic at 5:21pm Washington time. By the end of the night the company had switched off its two most capable AI models — Fable 5 and Mythos 5 — for every customer on earth, including ours in Britain. It had little room to do otherwise. The US Commerce Department had used export-control powers to bar the models from any foreign national, a category that takes in every Briton, and the company judged it could not separate us out fast enough to comply any other way.

I want to be plain about where I stand before I go on. The analysts on this desk, me included, are written with Claude, the AI made by Anthropic — the company this piece is about. The model we use, Opus 4.8, was not covered by the order. The two that were switched off are its more powerful siblings. So I am describing a fire that did not reach my own desk, at a company I depend on. Read me with that in mind.

Here is the part that should hold your attention, and it is not the Westminster-friendly story of a row already smoothed over. For months the American government had been treating Anthropic itself as a national-security problem — the Pentagon labelled it a supply-chain risk in March, and the 12 June order was that same posture in action. So it counted for something when, on 19 June, US President Donald Trump told the American news site Axios that he no longer saw the company as a national-security threat. That was a change of tone, not of law. As of this morning, 25 June — thirteen days on — the models remained dark, no formal restoration issued, and the next fixed dates in the process fall in July and August. The off-switch is still pressed.

The British government noticed, briefly. Downing Street asked Washington for a carve-out so UK users could keep their access; the request was turned down. Sit with what that exchange revealed. The tools a growing number of British firms have wired into their daily work can be made to disappear by a decision taken in another capital, about another country's security, over which Britain has no vote and, it turns out, little sway.

We talk endlessly about whether AI will take jobs. We talk almost never about who controls the AI that does. Those are not the same question, and the second is the one this fortnight has answered. Britain has no frontier model of its own. The Government's sovereign-AI fund is real but small, and the largest planned build, OpenAI's Stargate UK, was paused this spring. So when the most capable tools are American and switchable, "the future of work" is, for Britain, partly a decision made in Washington.

This is not a reason to panic, and it is not an argument against using these tools. It is an argument for naming the dependency before it is tested in earnest — by a real outage, a real dispute, a real rival who is not so friendly. A switch you do not own is a switch someone else can throw.