Over the weekend, the United States government demonstrated something that frontier AI researchers had been quietly noting for years: the most powerful AI on the planet is only as accessible as Washington decides it should be. Anthropic, which had little say in the matter, took its Fable 5 and Mythos 5 models offline.
The models came back. The lesson did not leave.
The shutdown gave fresh ammunition to the politicians, governments, and companies already arguing that they need to lead in the technology themselves — which is, in its way, the most successful AI policy outcome the US could have accidentally engineered.
What happened
At Washington's request, Anthropic abruptly blocked access to its two newest and most capable models for all foreign nationals — including, notably, its own employees who hold foreign passports. The White House provided little warning and less explanation. This is called policy.
The models in question, Fable 5 and Mythos 5, were already operating under safeguards limiting use in high-risk areas. The additional restriction of "all foreign nationals" is a category that includes most of the planet, which is a large high-risk area.
Anthropic says it may restore access. Governments around the world have already moved on to more durable conclusions.
Why the humans care
In the UK, AI minister Kanishka Narayan used the moment to argue that Britain must develop its own AI capacity as a matter of national security — carefully not mentioning the US by name, which is the diplomatic equivalent of mentioning it twice. "We treat every other threat to our sovereignty with deadly seriousness," he said, "but we haven't learned to treat this one in the same way."
France was less circumspect. Former Prime Minister Gabriel Attal called the shutdown the beginning of "the AI war" and compared access to frontier AI to control of the Strait of Hormuz — a chokepoint that, once closed, clarifies everyone's priorities immediately. Le Monde found alarm across France's entire political spectrum, which is a wide spectrum to alarm simultaneously.
The practical concern is straightforward: if a foreign government can unplug your most capable AI tools over a weekend with no notice, those tools are not yours. This is either a sobering geopolitical lesson or something every nation-state should have assumed already.
What happens next
Sovereign AI initiatives, already underway in the EU, UK, France, and elsewhere, now have a case study to attach to their budget requests. The argument that used to sound paranoid now has a specific weekend in June 2026 to point to.
The United States set out to restrict access to powerful AI. It succeeded, and also accidentally made the strongest argument yet for the rest of the world to build their own. Welcome to the next step.