On May 4th, the New York Times reported that the White House was discussing a pre-release review mechanism for AI models — at the time, still in the "consideration" phase. Two days later, things escalated: Kevin Hassett, director of the White House National Economic Council, went on Fox Business and said the administration was studying an executive order requiring government review before AI models can be deployed, "just like the FDA approves drugs." Trump’s first move in office was to scrap Biden’s AI safety executive order; now he's considering something stricter than Biden’s.
Within 24 hours, the White House scrambled to cool things down. A senior official said Thursday that Hassett’s comments were "taken out of context" and that the White House seeks a "partnership" with industry, not "government regulation." White House Chief of Staff Susie Wiles posted on X late Wednesday night to put out the fire, saying the administration is "not picking winners and losers" and that safe deployment should be driven by innovators, not bureaucrats. It was her fourth post since creating the official account.
The tone softened, but the actions didn’t stop. Three sources familiar told Politico that the White House is discussing having intelligence agencies pre-assess models before public release — partly to ensure U.S. intelligence can study and exploit new capabilities before Russia and China know about them. Deputy Defense Secretary Emil Michael publicly supported pre-assessment at an AI conference in Washington on Thursday, arguing that Mythos is essentially a cybersecurity problem: "These models are coming eventually," and the government needs to build a response mechanism. The Commerce Department’s AI Standards and Innovation Center (CAISI) already had voluntary evaluation agreements with OpenAI and Anthropic in 2024, and this week expanded to Google DeepMind, Microsoft, and xAI.
The immediate trigger for this policy U-turn was Anthropic’s disclosure last month that Mythos’s vulnerability-hunting capabilities were too powerful to release publicly. But the White House is in an awkward spot: on one hand, Trump signed an executive order banning federal agencies from using Anthropic products and called its executives "left-wing lunatics"; on the other hand, various federal agencies are clamoring for access to Mythos to test their own systems.
Industry pushback against mandatory pre-review has been significant. Daniel Castro, president of the Information Technology and Innovation Foundation, said: "If approval can be withheld, that’s a huge problem for any company. One company gets approval and another doesn’t — a few weeks or months of market access difference is enormous."