The White House now sits outside an unusual AI-safety coalition of rivals, allies and laboratories

At the UN General Assembly and during Xi Jinping’s Washington visit, the Trump administration rejected international efforts to constrain AI development. White House science adviser Michael Kratsios described global governance efforts as an intervention, while Trump argued that US companies should be left to accelerate. Xi, by contrast, called responsible development necessary, and Chinese officials supported a global framework.

The diplomatic alignment is unusual. European governments want common rules; China is at least rhetorically endorsing them; and OpenAI, Anthropic and Google are building voluntary coordination around testing and model risk. Frontier laboratories have commercial reasons to shape regulation in ways they can satisfy, so their support is not disinterested. Even so, they now stand closer to the UN position than the US executive branch does.

Domestic politics are also less uniform than the administration’s message. Republican senator Todd Young has asked the State Department to convene national-security meetings with cyber experts and developers on threats such as AI-enabled attacks on critical infrastructure. Polling cited by Semafor shows broad concern across party lines about jobs and rapid development. A race-with-China argument does not automatically produce public consent for having no ground rules.

The practical consequence is fragmentation. Without the home government of most frontier labs, international standards may remain voluntary and nonbinding. Companies will then negotiate their own test thresholds while competing to release, and China can present itself as the advocate of restraint without accepting intrusive verification. US leadership need not mean slowing every model; it could mean defining incident reporting, secure evaluation boundaries and shared measurements before a crisis makes those arrangements harder.