Trump Unblocks AI Development

“President Trump's June 2 executive order on artificial intelligence is less ambitious than one he withdrew without signing last month. The earlier order "could've been a blocker" to American AI development, he said. The new one is better, but the thinking behind it still rests on a dangerous confusion.

 

The signed order directs federal agencies to strengthen defenses against advanced AI systems and creates a voluntary framework for labs to share frontier models with the government before public release. It explicitly prohibits interpreting the executive order as mandating licensing or preclearance. The administration wants to gain visibility into frontier capabilities without slowing American labs and advantaging China.

 

That tension is real, and the president deserves credit for acknowledging it. But the entire debate conflates two quite different things.

 

Regulating the use of AI differs from regulating its development. The European Union's AI Act, U.S. state-level initiatives and sector-specific agency guidance govern how AI is used in hiring decisions, medical devices and autonomous systems. No quarrel here with that project.

 

Regulating the development of increasingly capable AI at the global frontier, however, runs into three walls Washington can't climb.

 

The first is verification. Unlike a nuclear-weapons program, an AI lab emits no radiation, handles no controlled materials, and leaves no atmospheric or seismic traces. Advanced models can be copied in seconds and transmitted digitally, but centrifuge cascades can't fit in a laptop. Weapons inspections are possible because there is something physical to inspect. There is no equivalent for AI. A bedroom or a bunker is a plausible laboratory in a way it has never been for previous dual-use technology.

 

The second wall is self-interest. On the day that nations sign arms control agreements, the costs of compliance appear bearable. But the Kyoto Protocol, Copenhagen Accord and Paris Climate Agreement all showed how quickly nations revise commitments when costs -- domestic, international or both -- become unbearable. Limits imposed on American labs but ignored by the Chinese aren't a safety policy but a strategic abdication.

 

The third wall is the breadth of AI's benefits. Arms-control agreements ask nations to forgo mass destruction. But restricting AI development asks them to forgo better cancer diagnostics, faster drug discovery, improved weather forecasting and productivity gains compounding across every sector. The constituencies for AI development are oncologists, farmers and educators, not defense contractors. It is hard to imagine a politician or party surviving while opposing such interests.

 

The signed executive order points toward some elements of a workable policy. Voluntary sharing of frontier models with government evaluators creates accountability without pretending to control global development. Hardening critical systems against AI-enabled attacks, building detection capabilities for AI-generated disinformation, and investing in research to understand better how AI systems reach their conclusions all reflect reasonable expectations for what policy can achieve.

 

But the order doesn't directly confront the core problem: Increasingly capable AI will be developed globally regardless of what Washington decides. The actors of greatest concern are state-sponsored programs outside any Western framework and the global community of open-source researchers. They are entirely beyond the reach of any executive order signed in Washington.

 

The president's instinct to pull the original order was correct. The challenge is to build on that decision by designing policy around who leads at the AI frontier, under what conditions of transparency, and with what resilience against misuse, rather than premised on controlling something that can't be controlled.

 

---

 

Mr. Koonin is a senior fellow at the Stanford University's Hoover Institution. He served as undersecretary of energy for science 2009-11.” [1]

 

1. Trump Unblocks AI Development. Koonin, Steven E.  Wall Street Journal, Eastern edition; New York, N.Y.. 10 June 2026: A15.