President Donald Trump has delayed signing an executive order that would have moved the U.S. government closer to pre-release security reviews for advanced AI models.
The order was expected to direct the Office of the National Cyber Director and other federal agencies to develop a process for evaluating certain AI systems before they reach the public. Trump said he objected to parts of the draft language, arguing that the United States should avoid rules that could slow its lead in artificial intelligence.
“I didn’t like certain aspects of it,” Trump told the White House press pool, adding that the U.S. is ahead of China and other competitors and that he did not want to get in the way of that position.
Why the AI order stalled
The central tension is whether the government should be able to examine powerful AI models before they are released, and how far in advance companies would need to provide access. One reported point of friction was a proposed requirement that AI companies share advanced models with the government between 14 and 90 days before launch.
For AI labs and enterprise buyers, that kind of review process would be a major policy shift. It could affect product timelines, security testing practices, and the way vendors prepare high-capability systems for public or commercial deployment. internal-link-topic: AI model security testing
Trump framed the issue as a competitiveness concern. He said the draft language “could have been a blocker,” suggesting the administration may still pursue AI security oversight but with narrower or revised wording.
What the order was meant to address
The anticipated order was tied to growing concern that more capable AI systems can accelerate offensive cybersecurity work. The issue has become more urgent as advanced models show stronger ability to find, explain, and potentially exploit software vulnerabilities.
The draft policy would not have been a final security regime on its own. Instead, it would have instructed federal officials to build a process for evaluating models before release. That process could have shaped how companies document risk, cooperate with agencies, and demonstrate safeguards for highly capable systems.
For security teams watching the policy debate, the practical question is whether future U.S. rules will focus on voluntary testing, mandatory pre-release access, or a hybrid approach that applies only to the most powerful models. internal-link-topic: AI regulation for cybersecurity
What happens next
The delay does not necessarily mean the idea is dead. It does, however, show how difficult AI security policy becomes when national security concerns collide with commercial speed and geopolitical competition.
If the administration revises the order, the most important details will be the scope of covered models, the timing of any review requirement, and whether companies must hand over model access before launch. Those choices will determine whether the policy becomes a narrow security checkpoint or a broader gatekeeping system for frontier AI releases.