White House distances itself from tighter AI regulation

The Trump administration spent the week sending mixed signals about AI regulation, leaving tech companies uncertain about whether the White House is about to impose significant new oversight on the industry or stick to the light-touch approach the president promised during his campaign.

The confusion started Tuesday when POLITICO reported that the White House was considering a vetting system that could require AI companies like OpenAI, Anthropic, and Google to submit their models for government review before releasing them to the public. On Wednesday, Kevin Hassett, director of the White House National Economic Council, told Fox Business that the administration was studying an executive order requiring pre-release safety testing comparable to the FDA's drug approval process.

That comparison to the FDA — an agency known for lengthy, expensive approval processes — set off alarm bells across Silicon Valley. If the government can withhold approval before an AI model goes to market, the competitive implications are enormous. Daniel Castro, president of the Information Technology and Innovation Foundation, put it bluntly: "If one competitor got approval and the other didn't, that could have a huge impact if we're talking weeks or months in terms of their market access."

The backlash was swift enough to trigger a rare public statement from White House chief of staff Susie Wiles, who posted on X Wednesday night that the government is "not in the business of picking winners and losers." It was only her fourth post since creating her official account. A senior White House official told reporters Thursday that Hassett's remarks had been "taken out of context a little bit" and that the White House is looking for "partnership" with companies rather than pursuing "government regulation."

But the underlying concern driving the regulatory discussion is real and pressing. Anthropic's Claude Mythos, announced last month, demonstrated AI capabilities so powerful at finding and exploiting software vulnerabilities that the company chose not to release it publicly. Instead, it shared the model exclusively with a few dozen tech and security organizations so they could patch critical systems before malicious hackers could exploit the same capabilities. Anthropic has predicted that similar capabilities will be widely available within a year.

OpenAI followed suit Thursday evening, announcing limited previews of GPT-5.5-Cyber, its own vulnerability-finding tool. The arms race in AI cybersecurity capabilities is accelerating faster than most policymakers anticipated, and the government is scrambling to figure out how to keep up.

The national security angle is also driving the push for oversight. According to three people familiar with the plans, the White House is discussing tapping the intelligence community to pre-assess AI models and exploit their capabilities before adversaries like Russia and China learn about them. Defense Undersecretary Emil Michael framed the challenge directly at an AI conference Thursday: "The Mythos moment is really a cyber moment; how is the U.S. government going to deal with cyber? These models are coming one way or the other."

The regulatory discussion is further complicated by the ongoing feud between the administration and Anthropic. After Anthropic refused to allow the Pentagon to use its models in autonomous lethal attacks and mass surveillance, Defense Secretary Pete Hegseth designated the company a supply chain risk — an unprecedented move that barred Anthropic's models from use on Defense Department contracts. Trump separately ordered federal agencies to stop using Anthropic's products. Yet federal agencies have been pushing to get access to Mythos precisely because of its extraordinary security capabilities.

The tension between innovation and security is not new, but the speed of AI development has compressed the timeline for resolving it. The administration that campaigned on slashing AI regulation now finds itself confronting the possibility that the very models it encouraged American companies to build may require exactly the kind of oversight it promised to eliminate.

**What This Means For You:** If you work in AI or tech, this week's regulatory whiplash means policy risk is now a first-order concern for your company's strategy — model release timelines, compliance costs, and competitive dynamics could all shift depending on what the executive order actually says. If you're in cybersecurity, the government's interest in pre-release access to AI models creates potential partnership opportunities, but also raises questions about whether your defensive tools will be available to you before they're available to intelligence agencies. If you're simply watching this as a citizen, the core tension is clear: the same AI capabilities that could protect critical infrastructure from cyberattacks could also be weaponized by adversaries, and the government is trying to figure out how to manage that risk without killing the industry that created it. The FDA comparison may have been walked back, but the problem it was trying to solve hasn't gone anywhere.