On Thursday afternoon at the White House, the signing table was set. Representatives from OpenAI, Anthropic, and Google were seated in the audience. But the document was never signed.
Trump later explained:
"I called it off because I didn't like certain aspects of it."
What the Executive Order Would Have Done
The core provision of the draft was simple: AI models must be submitted to federal officials for review 90 days before release. It was modeled after the UK's AISI (AI Safety Institute) approach, with multiple agencies overseeing pre-market reviews of frontier models to ensure no major risks before deployment. OpenAI, Anthropic, and Google had previously discussed details with the Hassett team.
According to White House national security officials, the measure was necessary: a checkpoint before any model with significant cyberattack or biological risk capabilities could be released to the public. But Trump changed his mind this week.
Who Pushed Back
David Sacks, the former White House AI and crypto advisor who left his formal role six months ago but maintains close ties with the tech industry, was key. The tech sector's core argument to him: The US is racing China in AI, and a 90-day government review would give the opponent an edge.
Anthropic itself noted in a recent policy paper that the US frontier model lead over China is "only a few months." Adding a mandatory 90-day review window would essentially erase that lead. OpenAI spent $1.56 million on lobbying in Q1, as we reported earlier, much of it focused on such critical policy moments. Sacks brought these concerns directly to Trump.
Trump's Reasoning
Hours before the scheduled signing, Trump explained his reversal to reporters: the AI industry is "bringing huge benefits," the US currently leads China and "cannot afford to lose that advantage," and the executive order "could become a blocker." His calculus was clear: better no regulation than adding an extra step for US AI companies compared to Chinese rivals.
Months of Preparation Wasted
Looking at the timeline: In January, Trump scrapped Biden-era AI safety assessment requirements. In March, the White House issued a federal AI blueprint emphasizing state law preemption and training copyright exemptions. In early May, economic advisor Hassett signaled that AI models would face FDA-style pre-approval for each new version, with an executive order "within two weeks." On May 21 (Thursday), the signing table was set—then Trump called it off. In just four months, policy swung from regulation to deregulation and back.
What's Next
The draft executive order remains on the table, with no new signing date. Internal White House divisions persist: the national security faction wants NSA involvement, mandatory reviews, and strict AI safety controls; the innovation faction (led by Sacks) prefers voluntary frameworks and low barriers to avoid ceding ground to China. Until these sides reconcile, the next signing ceremony remains uncertain.
Meanwhile, state-level bills in California and New York are advancing. If the federal executive order remains stalled, state AI regulation may arrive first, potentially forcing the White House to revisit the preemption debate.
Conclusion
The signing table set and then removed marks a temporary win for Anthropic, OpenAI, and Google. But they understand that whether the White House regulates AI ultimately depends on whether the US is willing to pay a small speed cost for AI safety. Trump's answer this time was no. The next time a draft lands on his desk, the question will be: how long can this line hold?
Sources: Trump Postpones AI Executive Order Signing (CNBC); White House Postpones Executive Order on AI (CNN); CocoLoop; Trump Delays AI Executive Order on Pre-Release Model Reviews (Yahoo Finance)