Trump Pulled the AI Order Because Nobody Wants to Slow the Race
President Trump called off an expected White House signing ceremony for a new artificial-intelligence executive order hours before it was scheduled, saying he did not want anything to interfere with America’s lead over China. AP reported that the order would have created a voluntary framework for reviewing the national-security risks of advanced AI systems before public release, with major U.S.-based firms such as Anthropic, OpenAI and Google involved. The delay follows rising concern from finance and security officials about frontier models’ ability to find cybersecurity vulnerabilities. Treasury Secretary Scott Bessent and outgoing Fed Chair Jerome Powell had convened Wall Street executives in April to discuss risks around Anthropic’s Claude Mythos model. The political fault line is now clear: Washington wants to look vigilant about AI danger, but the same officials fear that any real friction could hand advantage to China or slow the companies already becoming national infrastructure.
The interesting part of this story is not that Washington changed its mind. Washington changes its mind every time a powerful industry finds the pressure point. The interesting part is how quickly artificial intelligence has moved from “technology product” to “strategic asset.” Once a model can plausibly affect bank cybersecurity, software vulnerabilities, military planning, and the balance with China, the normal consumer-tech playbook stops working.
That is why the pulled executive order matters. A voluntary pre-release review sounds mild. It is not a nationalization plan. It is not a ban. It is not even necessarily a hard licensing regime. But in the current AI race, even a soft checkpoint becomes a signal. If the White House says frontier models should be vetted before release, investors hear delay risk, companies hear regulatory discovery, and foreign competitors hear that the United States may be willing to slow its own champions.
This is the incentive problem nobody wants to say plainly. The people who understand the risks also fear the consequences of acting on them. Treasury officials can warn bankers that a new model may change the cybersecurity equation, but the political system still treats speed as national security. The companies want the credibility of partnership without the constraint of permission. The government wants the aura of oversight without being blamed if America loses technological ground. The result is a familiar compromise: announce seriousness, keep the pipeline moving, and hope the bad scenario stays theoretical.
For ordinary citizens, this is not an abstract Silicon Valley fight. If frontier AI becomes embedded in banking, government services, surveillance, law enforcement, insurance, hiring, and military systems, the public will live inside decisions made by tools it cannot audit and institutions it already struggles to trust. The same government that cannot reliably explain how data brokers track sensitive locations now has to decide when a private model is too powerful to release. That should make people nervous.
The China frame is real, but it can also become a shield. Every industry with enough political leverage eventually learns to translate its own growth strategy into national-security language. Data centers become strategic infrastructure. Chips become industrial policy. Models become deterrence. At some point, “we cannot slow down because China” becomes the all-purpose answer to every safety question, every labor question, every privacy question, and every accountability question.
The better question is not whether America should lead in AI. Of course it should. The question is what kind of lead it is building. A lead based on resilient systems, clear liability, transparent standards and public confidence is different from a lead based on permanent emergency logic. The second one may move faster, but it also concentrates power in the least accountable places: inside private labs, federal agencies, closed security briefings and financial institutions that get warnings before the public gets explanations.
Trump’s hesitation shows the real bargain being written. The government may talk about guardrails, but the first-order incentive is still acceleration. The companies know it. The markets know it. China hawks know it. Citizens should know it too. When the most powerful technology firms can make caution look like surrender, oversight becomes theater unless someone is willing to define the rules before the next crisis defines them for us.