A planned AI security order has been paused at the White House
President Donald Trump has delayed signing an executive order that would have created a process for the U.S. government to evaluate advanced artificial intelligence models before they are released. The decision leaves a significant federal AI security proposal in limbo and underscores the administration’s effort to balance competitive pressure with rising concern over dual-use model capabilities.
According to the White House press pool remarks cited in the source report, Trump said he was unhappy with parts of the order’s language and did not want to do anything that could interfere with U.S. leadership in AI. His comments framed the issue in geopolitical terms, with competitiveness against China presented as a key reason for caution.
The expected order would have tasked the Office of the National Cyber Director and other agencies with developing a government review process focused on security risks before advanced models reached the public. That would have marked one of the clearest recent attempts to formalize pre-release scrutiny for powerful AI systems at the federal level.
The proposal appears to have centered on early access for government review
One of the main sticking points, according to the report, was proposed language that would require AI companies to share advanced models with the government between 14 and 90 days before launch. Trump reportedly worried that this aspect of the order could become a blocker.
That concern goes to the core of the present AI policy debate. Pre-release evaluation is one of the few tools policymakers have for addressing cybersecurity and misuse risks before a model is widely deployed. But mandatory early sharing also raises industry concerns over speed, confidentiality, and whether regulatory friction could slow domestic firms in a market defined by rapid iteration.
The delayed order was partly motivated by fears surrounding models that can identify and exploit vulnerabilities quickly. The source text specifically points to concerns following the release of Anthropic’s Mythos and OpenAI’s GPT-5.5 Cyber, both described as capable of rapidly finding and exploiting security flaws. In that context, the administration appears to have been weighing the risk of harmful capabilities against the strategic cost of placing new obligations on U.S. developers.
The delay highlights the tension between safety oversight and industrial policy
What makes this development consequential is not only the pause itself, but what it reveals about Washington’s current stance. Rather than moving immediately toward stricter pre-release controls, the administration is signaling that any such mechanism must be compatible with its competitiveness agenda.
That does not necessarily mean the order is dead. Trump’s remarks focused on dissatisfaction with the language, not a categorical rejection of government evaluation. But the pause indicates that the White House is still negotiating the boundary between security review and industrial restraint. In practical terms, that boundary will determine whether any future system is advisory, mandatory, narrow, or broad.
The report also notes an unofficial explanation circulating in media coverage: that not enough technology chief executives could get to Washington on short notice. If accurate, that would suggest the administration wanted a higher-profile launch involving industry leaders, reinforcing the extent to which AI policy is now being staged not just as regulation but as a public alignment exercise between government and major firms.
Why the decision matters now
The delay comes at a time when advanced model capabilities are pushing AI governance toward more concrete questions. General commitments to safety are no longer the only issue. Policymakers are now confronting operational matters such as when companies must disclose models, who gets access before release, and what kind of testing is credible enough to justify deployment.
Those questions become more urgent when the models in question can support cyber offense, automate vulnerability discovery, or lower the expertise needed for harmful use. A pre-release review framework is one possible response, but it is also one of the most politically fraught because it directly touches product timelines and trade secrets.
Trump’s decision therefore serves as a marker of where the federal debate currently sits: advanced AI is widely viewed as a national capability issue, and any security mechanism that appears to slow industry may face resistance, even if the underlying risks are acknowledged.
What to watch next
The next phase will likely turn on whether the administration rewrites the order into something more limited or seeks another route for coordination. A narrower framework could emerge, focused on specific capability thresholds or voluntary cooperation rather than broad advance-sharing requirements. Another possibility is that agencies continue developing evaluation processes without an immediate high-profile executive action.
For companies building frontier systems, the episode is a reminder that U.S. AI policy remains unsettled at the point where national security, cyber risk, and global market leadership intersect. For policymakers, it shows how difficult it is to impose meaningful oversight while also promising that America will move faster than its rivals.
The order’s delay does not settle that conflict. It makes it more visible.
This article is based on reporting by TechCrunch. Read the original article.
Originally published on techcrunch.com
