I joked the other day about how long it would be before we all have to fill out a long KYC government form — pledging allegiance to the United States — just to keep using their AI models. People laughed.

Then I read the news, and the joke stopped being funny.

On 12 June, a directive from the US Commerce Secretary forced Anthropic to disable Fable 5 and Mythos 5 worldwide. Not throttle them. Not license them. Switch them off — for everyone, including American citizens — because the rules could not cleanly separate who was allowed to use them and who was not.

The stack has been climbing for years

That did not come from nowhere. For four years, US export controls have been climbing the AI stack one rung at a time. First the chips. Then the manufacturing equipment that makes the chips. Then raw compute and cloud access, with Know Your Customer checks, nationality screening and remote-access controls written straight into the licence conditions. Now the models themselves.

Each step looked reasonable on its own. Stacked together they describe a single trajectory: capability that used to be a product you bought is quietly becoming a permission you are granted.

Capability that used to be a product you bought is becoming a permission you are granted.

The trusted-partner trap

At the G7, the proposed fix was “trusted partner” access for allies. Australia sits on that short list, so notionally I am on the right side of the fence.

That is exactly what worries me. A place on a list is not the same as owning the thing. It is access — conditional, revocable, and dressed up as security. The fence is comforting right up until you remember you are standing on the wrong side of the gate, and someone else holds the key.

The world Elysium imagined

This is the world the film Elysium imagined. The people with the right credentials live above everyone else, with access to capability the surface is denied. Swap the orbital station for an API endpoint and the picture gets uncomfortably close.

Access that is granted can be withdrawn. A place on the trusted list is not a right, it is a privilege — and privileges move with politics. They last exactly as long as the relationship that issued them, and not a day longer. You do not get a vote on the foreign policy your business now quietly depends on.

The case for sovereignty

For me, this is the clearest argument yet for technological sovereignty. Not isolation. Not rebuilding everything from sand up. But capability you actually own, and a deliberate refusal to wire your business so it depends entirely on a single gatekeeper an ocean away.

Sovereignty is a spectrum, not a switch. Open-weight models you can run yourself. A second provider in another jurisdiction. Data and prompts you can pick up and move. The ability to keep operating — degraded but alive — on the day the permission slip is revoked. None of that is paranoia. It is just declining to put your whole operation on someone else’s switch.

The question worth sitting with

A few weeks ago I wrote that the real question isn’t whether you use AI, but what you have pointed it at. Here is its companion. It isn’t only what your AI optimises for. It’s who holds the off switch.

Optimise with AI. Just be careful you are not quietly optimising your whole operation around someone else’s permission slip.

Nigel Price is the founder of Digital Discovery Group, specialising in ecommerce strategy, digital transformation, AI-powered platforms, and managed cybersecurity services for small and medium businesses.