The government does not have to say, “You may not speak.” It can say, “Your license is under review.” It can say, “Your funding is gone.” Now it can say, “Your model is available only to approved users.”
That is how information control works in a modern administrative state: not always through one giant ban, but through pressure on the channels through which knowledge flows.
The administration’s stated target is China. That concern is not frivolous. China wants advanced AI for military, intelligence, cyber and industrial advantage. No responsible government would ignore that.
But the China argument depends on a premise that is becoming less true by the month: that frontier AI is an American secret China cannot get unless we hand it over. China is no longer outside the gate looking in. Chinese models from DeepSeek, Alibaba, Moonshot, Zhipu and MiniMax are increasingly capable. The United States may still lead at the top. But the gap is not the gap between having nuclear weapons and not having nuclear weapons. It is a narrowing software race.⁶
That changes the logic of export control. It works best when the controlled item is scarce, physical and hard to reproduce. It works less well when the controlled item is software, training know-how, model behavior and capabilities that can be approximated, distilled or rebuilt.
The people most reliably stopped by a U.S. access restriction are not China’s military or intelligence services. They are lawful users: American developers, independent researchers, universities, startups, hospitals, cybersecurity teams, allies and ordinary businesses that follow the rules. If the best American models become politically rationed, the world will not stop using AI. It will look for AI that Washington cannot turn off.
At the VivaTech conference in Paris, the word of the week was sovereignty. Reuters reported that executives from Siemens, Renault, Orange and ChapsVision said they already use a mix of U.S., Chinese and European models to avoid depending on any one provider. The Anthropic order gave that fear a body. A French company, a German manufacturer or a Spanish defense contractor could build around an American model and discover that access depends on a letter from Washington.⁷
That is not just a vendor risk. It is a sovereignty risk.
If advanced American AI is available only by Washington’s grace, allies will build around that risk. Europe will accelerate its own models, sovereign cloud and open-weight alternatives. Canada will discover that its hydro power, cold geography and AI research base are strategic assets. The more America turns AI into a permission system, the more it teaches its allies to build systems America cannot turn off.
Trump says he is protecting America’s AI lead. He may instead become the best recruiter Europe’s AI industry ever had.