The U.S. just invented AI export licensing without calling it that, and nobody seems to have noticed they're now gatekeepers to the world's most powerful models.
The Summary
- The White House lifted export restrictions on Anthropic's Mythos and Fable models weeks after forcing the company to suspend foreign national access
- This creates a de facto licensing regime for frontier AI that the administration won't acknowledge exists
- Every AI lab building at the frontier now operates at the pleasure of an executive branch that can flip access on and off with no clear rules
- The precedent matters more than the policy: we now have permission-based AI deployment in America
The Signal
The Trump administration ordered Anthropic to cut off foreign nationals from its most advanced models, then reversed course and lifted those restrictions weeks later. On paper, this looks like policy whiplash. In practice, it's something more important: the birth of America's first real AI licensing regime, built in the shadows of export control law.
Fortune notes the administration has created licensing requirements without admitting that's what they're doing. No public framework. No clear thresholds. No appeals process. Just case-by-case decisions from bureaucrats who can say yes or no to your model reaching 95% of the world's population.
"The trajectory of frontier AI development is now subject to executive whim dressed up as export controls."
This matters because every AI lab racing toward AGI just learned they're building on rented land. You can spend $100 million training a model, but whether you can actually deploy it globally depends on a phone call from the Commerce Department. Anthropic got the green light this time. What about the next model? What about the next company?
The opacity is the point. By using export controls instead of explicit AI regulation, the administration gets three things:
- Unilateral authority with no Congressional oversight
- The ability to claim they're not regulating AI domestically
- A tool that works on both U.S. companies and foreign actors
The foreign national access ban was always going to be temporary. It was a shot across the bow. The real play was establishing that the government can reach into a private company's API and flip the switch. They did it once. They'll do it again.
The Implication
If you're building frontier models, your compliance team just became more important than your research team. The U.S. now has the ability to kneecap any AI company's international business overnight, and they've proven they'll use it. Start building relationships with Commerce Department officials now, because informal guidance is the only guidance you're getting.
For everyone else, watch which models get the green light and which don't. The pattern will tell you more about AI policy priorities than any white paper. We're not getting a formal licensing regime because we already have one. It's just dressed up as national security.