America Is Locking the World Out of Its Best AI
Two of the best models on earth, both behind a government gate now.
Anthropic put out Fable 5 a couple of weeks ago. The best model they’ve ever made. Within days, it was gone, and not because anything went wrong with it. The US government told them to pull it.
If you’re not a US citizen, you don’t get to use it. Living in San Francisco doesn’t help. Working at Anthropic doesn’t help either, they had to cut off their own foreign staff.
I live in Germany, so that includes me. Probably many of you reading this too.
The cutoff
What happened, roughly. In mid-June the Commerce Department hit Anthropic with export controls and ordered it to block Fable 5 and its bigger sibling Mythos 5 for every foreign national, inside the country and out. Anthropic can’t really tell who’s a citizen from an API call, so it just switched both models off for everyone. Worldwide.
The official reason was national security, and that was about all they said out loud. The reporting filled in the rest. A group linked to China had apparently gotten into the new model, and officials were nervous that Fable 5 could be talked into helping with real cyberattacks, the kind that go after banks and their ancient backend systems. Anthropic argued that one narrow jailbreak shouldn’t kill an entire commercial product. The government didn’t agree.
The logic is what the US already does with Nvidia and AMD chips. Some countries don’t get the good silicon. Now they don’t get the good software either.
OpenAI Too
And it isn’t only Anthropic.
OpenAI’s newest, GPT-5.6, got the same thing in a softer wrapper. The administration asked OpenAI to hold back the wide release and keep it to around 20 partners the government had vetted.
Sam Altman reportedly told staff that access gets approved one customer at a time, by Washington. So it’s not a blanket foreigner ban like Anthropic’s. It’s a guest list. And you’re not on it.
Different method, same result. The two best models in the world right now both sit behind a gate in Washington.
This switch is scary
A few American companies build the best models on earth, and the American government decides who gets near them.
That’s a scary and powerful switch, and unfortunately the lunatic in DC has his hand on it.
We’ve been here before with other things. Musk cut Starlink coverage over a war zone when it suited him, because he could. Europe got nervous enough about American payment rails that it’s now building its own digital euro.
Now it’s AI’s turn.
Which is, in some sense, worse, because AI is becoming the thing everything else runs on top of. Everyone in the world.
We need our own
Europe needs its own. So do other parts of the word. But we’re freaking far away.
AI models are not a nice-to-have anymore. Real labs and real models, built in the EU and elsewhere, that no foreign government can reach over and switch off.
We’ve got the start of one. Mistral, in France.
Their whole thing is open weights, models you can download and run on your own machines, which does matter. You can’t recall a model that’s already sitting on someone’s servers in Frankfurt. The rest of Europe’s AI scene is there too, just small… very small.
Aleph Alpha in Germany gave up chasing the frontier a while back. Black Forest Labs does image models.
It’s a start. Not much more than that. We are very far from having great AI models for everyday users.
The money is one thing
Just look at the money.
OpenAI is pulling in something in the mid-twenty-billions a year. Anthropic is past a thirty-billion run rate. Mistral expects around one billion euros this year. Round those however you like, they’re not in the same stratosphere.
The models say the same thing. Mistral’s best is the weakest reasoner of the frontier bunch. It’s okay. It can do a lot. But it isn’t the best by any means, and it isn’t pretending to be.
Then it gets circular… unfortunately.
You can’t train a top model without a huge pile of chips, the best chips are American, and the US decides who gets them. The same rule that just locked us out of Fable also makes it harder for us to ever build our own.
We’re behind, and a big reason we stay behind is the country we’re trying to catch.
See the problem…
To be fair
Some of this is a real concern.
If a hostile group did get into the model, and it really could speed up an attack on a bank, I understand hitting the brakes. Not every national-security line is a cover story.
There’s also a contradiction the US hasn’t sorted out. Trump’s AI pitch is that the world should run on American models, American AI everywhere as the default. You don’t get there by slamming the door on your best ones.
Every time Washington yanks a model, Mistral’s boring open-weight approach will get more attractive. A model you host yourself can’t be switched off by anyone.
The Bottom Line
Winning the AI race, is not in the cards for us in the EU. Maybe ever.
Just having something nobody across an ocean can turn off on a Tuesday should be the goal, though.
We don’t have that yet. We’ve got a small start, a long road, and a chip supply issue.
I’d love to work with a European AI model. Can’t right now. Not really.