When your AI vendor can be switched off by a government

Yesterday the US government ordered Anthropic to cut off access to two of its most capable AI models, Fable 5 and Mythos 5, under a national-security export-control directive. The order did not stop at America's borders. It reached every customer worldwide, and even Anthropic's own foreign-national staff. Anthropic complied within hours, because it had no real choice.

To its credit, Anthropic has said publicly that it disagrees with the order and is working to reverse it in the open. But step back from the specifics, because who was right matters less than what the episode reveals.

If your product depends on a frontier AI model run by a US company, your access to that model is subject to US government policy. Not your contract, not your SLA, not your vendor's good intentions. A directive citing national security can switch it off for everyone outside America overnight, and the best your vendor can do is apologise and appeal.

For a UK business that is not a privacy question. It is a continuity question. What happens to your product, your customers and your roadmap on the morning your core AI capability is revoked by a government you do not vote for and cannot petition?

This is the part of "sovereignty" that people tend to skip. It is not only about where your data sits. It is about whether the thing you built on can be taken away from you.

Pendra runs open-weight models on UK infrastructure. The weights sit on hardware that you or we control. No foreign directive can reach in and switch them off, because there is no foreign vendor in the loop holding the keys.

Build on something that cannot be turned off from the outside. If you want to talk through what that looks like for your product, get in touch, or browse the open models you can run on Pendra today.