Washington switched off the world's most powerful public AI. For everyone.
On Friday evening, a directive from the US government landed on Anthropic's desk. By the weekend, Claude Fable 5 and Claude Mythos 5 — the most capable AI models ever released to the public, four days old — were switched off. Not throttled. Not put behind a waitlist. Off. For everyone, everywhere.
What actually happened
The order, issued under US export-control and national-security authority, told Anthropic to suspend all access to the two models by any foreign national, whether inside or outside the United States. Because there is no clean way to fence "foreign nationals" out of a product used by hundreds of millions of people, Anthropic disabled the models for all its customers to comply. The stated trigger was a reported method of "jailbreaking" the models' safety system.
To its credit, Anthropic did not go quietly. In its public statement the company said it disagrees "that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," and that it is working to restore access as soon as possible. We take them at their word, and we expect the models to come back.
Why this matters even though it's temporary
Here is the part that matters for anyone doing confidential professional work, and it has nothing to do with Anthropic being careless. Anthropic is one of the most safety-obsessed labs on earth. It objected to the order. It complied anyway — overnight, worldwide — because it is a US company and the order came from the US government.
That is the whole lesson in one line:
When your AI vendor lives under another country's jurisdiction, that government — not your vendor, and not you — has the final say over whether you can use it.
A vendor's goodwill doesn't override that. Nor does its Data Processing Agreement, its "UK data region," its safety record, or how much you like the product. We've argued for a year that a US provider can be compelled — usually we're talking about the CLOUD Act and your data. This week made the same point in a starker way: the compulsion ran all the way to switching the product off.
"Any foreign national" means you
Read the directive again: any foreign national. If you're a GP in Leeds, a solicitor in Bristol, an accountant in Dublin, the United States considers you a foreign national. A tool built into a professional's working day was switched off by a government they don't vote for, for reasons they were never party to, with a weekend's notice.
This time the model was four days old and few people had come to depend on it yet. That's luck, not reassurance. The principle it demonstrated — that a foreign state can and will reach into a commercial AI and revoke access for everyone outside its borders — is the one that matters. Next time it could be a model your practice has built a month of habits around.
What this looks like in practice
Scenario A — AI under a foreign jurisdiction
Your drafting tool runs on a US company's model. A directive arrives from that company's government. The provider has no choice; it complies. Your access disappears on someone else's timetable, and you find out from the news.
Scenario B — AI on hardware a UK company owns
The same directive is issued. There is no US entity in the chain to serve it on, and no foreign authority that can order the service off. Nothing happens. You keep working.
Why a foreign order can't switch Hush off
Hush runs on hardware a UK company owns, in England. No US parent. No US cloud. No US API anywhere in the chain — we checked, because we built it. That means there is no US entity for a US directive to be served on, and nothing for a foreign government to switch off. The same structure that keeps your data outside the reach of the CLOUD Act keeps your access outside the reach of Washington. This isn't a feature we shipped last week in response to the news; it's been the architecture from day one.
Our whole pitch is an AI that minds its own business. As it turns out, that cuts both ways: ours minds yours, and no other country's government can make it mind theirs.
The honest caveat
We're not going to tell you Hush matches a frontier lab on raw capability. Fable 5 is an extraordinary model and we won't pretend otherwise. What we will tell you is that the AI you lean on for confidential work shouldn't be something a foreign government can revoke. For drafting under your review, on work you can't afford to lose access to or control of, "can't be switched off from abroad" beats "most powerful" every time.
Hush can't be switched off by a foreign directive.
UK company. UK-owned hardware. Zero US infrastructure. There's no one abroad to order it off — and we never read, sell, or train on your data. Try it on your real work, free for two weeks: no card, no calls, no auto-renewal.
Start your two weeks free → Run the CLOUD Act checkerSee also: Hush AI vs Claude — an honest comparison · Is your AI provider subject to the US CLOUD Act?
General information about the AI market, not legal advice. ICO Data Controller Registration: C1912355. Hush AI is founded and operated by a UK-registered doctor.