The Trump administration is drafting plans that would, for the first time under the current White House, subject frontier AI models to a government review before they reach the public. The New York Times reported the deliberations late on May 4, and the story rippled across Bloomberg, Axios, Tom's Hardware and Yahoo Finance through the morning of May 5. The proposal marks one of the sharpest reversals on AI policy since Trump rescinded the Biden-era executive order earlier in his term.
What's on the table
According to the reporting, the draft executive order would create a working group of technology executives and senior government officials tasked with studying — and potentially designing — review procedures for new AI models before public deployment. A second strand of the plan would give the federal government early or priority access to advanced models in advance of their commercial rollout. White House officials have already briefed leaders from Anthropic, Google and OpenAI on the contours of the package.
A White House official told the Times the chatter about a forthcoming order was 'speculation' and that any announcement would come directly from the President.
The Mythos catalyst
The trigger, by multiple accounts, is Anthropic's Claude Mythos. The company has publicly described Mythos as capable of finding zero-day software vulnerabilities at a scale and reliability that earlier models could not match, and it has declined to release the model broadly. Anthropic's Project Glasswing instead routes Mythos Preview to a narrow circle of critical infrastructure partners so defenders can patch issues before similar capabilities become commodity.
That posture — a frontier lab voluntarily withholding a model on national-security grounds — appears to have shifted internal calculations in Washington.
A different cast of advisers
The shift coincides with leadership changes inside the administration's tech orbit. David Sacks, the deregulation-minded AI czar, departed in March. Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent have taken a more active role in shaping AI policy in his absence, according to the reporting. Vice President JD Vance has previously cautioned that excessive regulation could 'kill' the industry; Trump himself has said the country can't 'stop it with foolish rules.' The new direction would test how durable those positions are when the technology in question is being characterized — by its own developer — as a cybersecurity inflection point.
Implications
If an executive order materializes in the form being discussed, the practical effect would be a return — at least in spirit — to pre-release safety testing requirements that the administration scrapped at the start of the term. The exact teeth of any regime, including whether reviews would be advisory or blocking, remain unsettled. Frontier developers should expect more questions from federal counterparts in the coming weeks regardless of whether the order is signed.
