Trump postpones AI executive order after Sacks raises industry concerns

TL;DR:

• US President Donald Trump postponed Thursday’s planned signing of an AI executive order, telling reporters he “didn’t like certain aspects of it” and citing competition with China; the order would have created a voluntary review process for frontier AI models prior to public release, with the NSA setting which systems counted as “covered frontier models”.
• The intervention was led by former Trump AI czar David Sacks, who called the president directly the morning of the planned ceremony after participating in the EO review earlier in the week; Semafor and Politico both reported that xAI’s Elon Musk and Meta’s Mark Zuckerberg also pushed back.
• Industry positions split publicly: OpenAI’s chief lobbyist Chris Lehane had been broadly supportive of “collaborating with the government on AI safety”; other industry executives had pressed for a 14-day rather than 90-day model-sharing window and for the intelligence community to lead review.

This is the clearest sign yet that the Trump administration’s AI agenda is being shaped in real time by the venture-capital and platform-CEO networks closest to the president, not by the federal-agency policy machinery. The order would have been the first US federal-level voluntary review regime for frontier models — a softer-touch version of the AISI-style evaluation pipeline that the UK and EU have built. Its postponement leaves a vacuum that EU and UK regulators will move to fill.

The pattern: industry lobbying killing a process industry helped design

The notable detail in Politico’s reporting is the timeline. Sacks took part in the EO review through the week and “White House officials believed he was generally happy with it and would support it” — until he raised concerns on Wednesday night and called the president Thursday morning. The order was not killed on substance over weeks of negotiation; it was killed in 12 hours by a single phone call. For UK and EU firms that have spent the last 18 months investing in voluntary AISI-style cooperation, the read-across is that US-domestic frontier-model oversight is now a much weaker constraint than it looked on Tuesday.

The split inside the industry matters separately. OpenAI’s public support reflects a calculation that being inside a voluntary process is cheaper than being outside an eventual mandatory one — the same logic that drove Anthropic’s open dialogue with the UK AISI and the now-defunct White House AI Safety Institute consultations. xAI, Meta and parts of the VC community appear to have made the opposite calculation. The Anthropic-OpenAI vs xAI-Meta divide on government cooperation is now structural, and the EO postponement makes it more visible.

Looking forward

Expect the EO to come back in revised form, likely with shorter review windows and reduced agency footprint. For UK government and AISI specifically, the postponement is a relative boost — UK frontier-model evaluation has a clearer regulatory mandate than US equivalents now have. For UK firms exporting AI services into the US market, the immediate uncertainty is whether the eventual rules will mirror the current draft (NSA-led review of covered models) or migrate further toward industry self-governance. Anthropic’s Mythos cybersecurity model — the kind of frontier capability the order was nominally designed to address — remains in commercial release, which sharpens rather than dulls the underlying policy question.