Trump AI executive order targets model cybersecurity review

President Donald Trump is expected to sign an executive order as soon as Thursday that would require cybersecurity reviews for advanced artificial intelligence models before release, according to people familiar with the planning cited by Bloomberg and CNN. The order follows warnings from intelligence officials that some frontier systems could be turned against U.S. networks and would be the first White House action on AI security since Trump took office.

The directive would set up a voluntary process for companies to submit advanced models for government review before launch and give the White House a direct role in those assessments, Reuters reported. The question is how far the administration is prepared to slow deployment to head off risks that officials and outside allies have spent months flagging.

The details are not fixed. One version of the order would give the government as long as 90 days to review covered models, Reuters reported. Some AI companies want a 14-day window instead, CNN said. The White House has not confirmed either figure. A spokesperson called any talk of policy specifics “speculation.”

Trump has asked technology executives to attend the signing, people familiar with the planning told Bloomberg. A public event would draw attention to a fight that has so far played out inside the administration and let the White House demonstrate it can move on AI without legislation from Congress.

Advocates of tighter oversight argue that company promises are not enough when models could plausibly be used for phishing, malware development or other cyber operations at scale.

“You can’t count on these people that are leading these AI companies to put our interests at heart and do what’s right to protect the American people,”
— Amy Kremer, quoted by Reuters

Kremer’s criticism captures a divide inside the Republican coalition. Many Trump allies have opposed AI regulation. But some of the same people say the largest model developers should not be left to police themselves when the stakes include military operations, intelligence collection and critical infrastructure.

Review window under debate

CNN reported that some companies want a 14-day review rather than 90 days. The gap matters. A longer window lets officials examine test results and decide whether a model poses cyber risk. A shorter one addresses a different worry: that drawn-out review would slow American firms while competitors abroad keep building.

Washington’s AI debate has been shaped by that tension for months. The government’s standards work already sits inside the National Institute of Standards and Technology’s Center for AI Standards and Innovation, or CAISI, which handles model testing and evaluation. The order would not resolve whether review should be mandatory. It would, however, give the administration something to cite while officials work out who runs the tests, what triggers them and how results get used.

Reuters reported that the order would fold model review into existing White House oversight instead of building a new regulatory apparatus. Trump can present it as a cybersecurity and national-security measure, sidestepping the political fight a broader AI bill would invite.

Security case and criticism

Officials arguing for the order say frontier AI no longer fits inside a technology-policy box. Systems that can speed up code generation, automate reconnaissance or help lower-skilled actors test computer defenses belong inside the government’s cyber-defense planning, they argue.

Critics say the administration risks overselling what pre-release review can accomplish if the same capabilities are already spreading globally.

“Holding back new AI models while the federal government vets them may allow the U.S. to gain a short-term advantage over adversaries but will not keep the technology out of enemy hands in the long term,”
— Neil Chilson, quoted by Reuters

The criticism exposes a gap the order does not close. The White House can slow or shape how American firms launch top-tier models. It cannot stop competitors abroad from building similar systems. The administration is left trying to show it is acting on AI security while keeping the process voluntary enough to avoid a fight with industry and declining to claim powers the order does not create.

Trump could sign within hours, making it among the first AI governance actions of his second term. Whether the order becomes a narrow review process for a handful of systems, or the start of a wider White House effort to write cybersecurity rules for frontier AI, is not yet settled.