Trump pulls AI executive order after adviser, tech CEOs revolt

WASHINGTON — President Trump pulled a planned executive order on artificial intelligence and cybersecurity hours before a scheduled White House signing ceremony on Thursday, after his top AI adviser and the chief executives of Meta and xAI intervened to kill it.

“I didn’t like certain aspects of it. I postponed it,” Trump told reporters, confirming the abrupt reversal that Axios first reported. The signing ceremony was to be a photo opportunity with technology CEOs flanking the president as he put the weight of the federal government behind new AI safeguards. By Thursday morning, the event had been scrapped entirely.

The order, drafted by White House chief of staff Susie Wiles and a small group of aides, would have required AI companies to voluntarily share advanced models with the government 90 days before public release. It also proposed a Food and Drug Administration-style approval process for frontier AI systems, a framework that National Economic Council director Kevin Hassett championed inside the administration as a way to vet the most powerful models before they reached consumers. Under the proposal, companies developing models above a certain capability threshold would submit their systems for federal review, similar to the pre-market clearance the FDA requires for new medical devices, according to Axios.

Months of infighting preceded the collapse. The proposed framework was developed in direct response to Anthropic’s Mythos model, which fewer than two months ago was demonstrated to autonomously discover thousands of severe cybersecurity vulnerabilities. The disclosure rattled corporate boardrooms and prompted the White House to move quickly on safeguards, with Vice President JD Vance saying the administration was prioritising data privacy in the wake of the Mythos disclosures.

But the order ran into fierce resistance from inside the administration and from the technology executives Trump had expected to stand beside him at the signing. David Sacks, the White House AI and cryptocurrency adviser, despised the order and told colleagues he considered it unnecessary, according to the Financial Times. Mark Zuckerberg, the Meta chief executive, and Elon Musk, who runs xAI, both spoke with Trump late Wednesday and early Thursday to argue that even a voluntary framework would handicap American AI companies as they compete with China.

“The whole thing was unnecessary — just something doomers wanted,” a source familiar with the discussions told Axios.

The episode exposed a defining split inside the Trump administration over how to regulate artificial intelligence. Sacks, Musk and Zuckerberg argued that any regulation, even a voluntary framework, would slow American AI firms as they race China. National Economic Council director Kevin Hassett, by contrast, had proposed the FDA-style review framework, and Treasury Secretary Scott Bessent was given an early look at Mythos as the administration weighed how aggressively to respond.

An industry official also questioned why the Treasury Department was tasked with a leading role in coordinating cybersecurity vulnerabilities, according to the Axios report. Cybersecurity coordination has historically sat with the Cybersecurity and Infrastructure Security Agency or the National Institute of Standards and Technology. The unusual assignment raised credibility concerns among technology policy observers, who argued that Treasury lacks the institutional expertise to evaluate model-level security threats.

“It could be CEOs, or egos in general. Everyone hates each other in the political tech space,” one government official told Axios.

The Mythos model has become the central exhibit in the argument over whether voluntary measures are sufficient. Semafor reported in mid-May that Mythos cracked macOS during internal testing and helped researchers uncover 271 vulnerabilities in the latest release of the Firefox browser. Security researchers have warned that voluntary testing regimes are unlikely to be adequate given those capabilities, which include the autonomous discovery of exploitable software flaws at a scale no human team could match. The American public appears to agree.

An Institute for Family Studies poll found that 82 per cent of Americans supported White House safety testing for advanced AI models, the Financial Times reported. The figure reflects broad anxiety about the national security implications of AI systems that can independently probe and breach software defences without human direction. Across the Atlantic, the European Union has already enacted binding AI rules through its AI Act, a contrast that some lawmakers in Washington have cited in calling for a more muscular federal framework.

The White House gave no timeline for when a revised executive order might appear. Senior officials indicated privately that the administration may attempt to salvage elements of the framework in a narrower form, though Sacks is expected to retain significant influence over any rewrite, according to the Financial Times. Trump’s public remarks on Thursday framed the decision squarely as a matter of American competitiveness.

“We’re leading China, we’re leading everybody, and I don’t want to do anything to get in the way of that lead,” Trump said.

For now, the security framework the White House spent weeks crafting has been shelved. Trump’s instinct to deregulate, reinforced by technology executives who have his ear and by an AI adviser who saw the order as the work of “doomers,” has left the administration’s AI policy rooted not in statute or agency rulemaking but in the arguments of the last phone call the president took.