Trump Halts AI Order Citing Risk to U.S. Lead Over China

President Trump on Thursday postponed signing an executive order on artificial intelligence, announcing the decision at a White House event where he also unveiled rollbacks of environmental rules for refrigeration equipment. He said parts of the draft could slow U.S. AI development and undermine the country’s lead over China.

The proposal, released earlier this month, would have established a voluntary framework for companies to provide the federal government early access to advanced, or “frontier,” AI models before public release. The framework aimed to let officials test models for national security and capability risks. It would also have allowed critical infrastructure providers, including banks, to receive pre-release access and included cybersecurity assessments to look for vulnerabilities in unreleased systems.

“I didn’t like certain aspects of it,” Trump told reporters in the Oval Office, adding that he was reluctant to impose anything that might hinder the U.S. position in AI. He declined to identify specific provisions he opposed or to say when the administration might return to the issue. He also said he discussed AI with Chinese President Xi Jinping during a meeting last week and that Xi acknowledged U.S. progress in the technology.

U.S. officials have increased scrutiny of advanced AI after tests showed some systems can identify software flaws and carry out complex cyber tasks. One model developed by Anthropic reportedly found hundreds of software vulnerabilities and demonstrated autonomous cyber operations during testing. Those results prompted consideration of greater review of powerful models prior to public release.

At the same time, major AI companies have expanded work with U.S. defense and intelligence agencies. Firms including OpenAI, Google and xAI have broadened partnerships with government bodies, and the National Security Agency has run advanced models on classified networks in some cases.

Administration officials said the voluntary review was intended to give the government time to test models for safety and security concerns while avoiding new binding federal controls. Some industry representatives warned that stringent pre-release requirements could slow innovation or advantage foreign competitors.

The White House has not released a revised version of the order and has not set a timetable for revisiting the proposal. Officials will continue discussions on how to balance testing and oversight with the goal of maintaining U.S. competitiveness in AI.