President Trump won’t be signing an executive order on AI security anytime soon. The order, which would have required AI companies to submit their models for government security reviews before release, has been indefinitely delayed after Trump said the language “could have been a blocker” for American AI development.

The move marks a significant reversal in the administration’s approach to AI regulation. The proposed order would have established a pre-release review process for frontier AI models, ostensibly to catch security vulnerabilities before deployment. But Trump told reporters he didn’t want to “get in the way of that leading,” referring to American dominance in AI development.

It’s a telling choice of words. The concern isn’t that pre-release reviews are unnecessary. It’s that they might slow things down.

What the order would have done

The details are still sparse, but the executive order appears to have been modeled loosely on pre-release security frameworks used in other high-risk industries. Companies developing large-scale AI models would submit them for government review, presumably to catch issues like exploitable vulnerabilities, potential dual-use capabilities, or other national security concerns.

The problem is that “pre-release review” can mean a lot of things. It could be a lightweight checklist process that takes days. Or it could be a months-long approval gauntlet that gives bureaucrats veto power over which models ship and when. Trump’s comments suggest he thought it looked more like the latter.

The phrase “could have been a blocker” is doing a lot of work here. Trump isn’t saying the order would have stopped AI development outright. He’s saying it might have. That uncertainty was apparently enough to kill it.

The political calculation

This isn’t the first time Trump has hesitated on AI regulation. After taking office, he quickly rolled back parts of Biden’s October 2023 executive order on AI safety, calling it “regulatory overreach.” But the security-focused order was different. It came from within his own administration, and it focused on national security rather than broader safety concerns.

So what changed? The most likely explanation is that Trump got pushback from the AI industry. Companies like OpenAI, Anthropic, and Google have been lobbying hard against pre-release review requirements, arguing that they’d hand China a competitive advantage by slowing American labs while Chinese companies operate without similar constraints.

That argument clearly resonated. Trump’s comments about not wanting to block American leadership suggest he bought the industry’s framing. Whether that framing is accurate is another question.

What happens next

The delay leaves the US without a clear framework for AI security oversight at the federal level. States are starting to fill the gap. California’s SB 1047, which would have required safety testing for large AI models, was vetoed by Governor Newsom last year, but similar bills are circulating in other state legislatures. Without federal action, we’re likely to see a patchwork of state-level requirements that could be even more cumbersome for AI companies than a single federal standard.

Congress isn’t moving much faster. There’s bipartisan interest in AI regulation, but no consensus on what it should look like. Some members want pre-deployment testing requirements. Others favor liability rules that punish harms after the fact. Still others think the market should sort it out.

The executive order delay also raises questions about how seriously the administration takes AI security risks. The national security establishment has been sounding alarms about AI for years. The concern isn’t just that bad actors might misuse AI systems, but that the systems themselves might be vulnerable to exploitation or manipulation in ways that create strategic risks.

Trump’s decision suggests those concerns take a back seat to maintaining American competitive advantage. That’s a defensible position, but it’s worth being clear about the tradeoff. Moving fast means accepting more risk. The question is whether we’re accepting it deliberately or just hoping nothing goes wrong.

For now, AI companies can breathe easier. They won’t have to run their models through a government review process before launch. But the reprieve might be temporary. If something does go wrong with a major AI system, expect the conversation about pre-release reviews to come roaring back.