Trump Shelves AI Safety Order After Industry Pressure

President Trump pulled back from signing a much-trailed executive order on artificial intelligence after a flurry of last-minute calls from technology executives, according to reporting in The Washington Post and Axios. The signing had been scheduled for Thursday.

The draft order would have set up a voluntary arrangement under which leading AI developers handed over advanced models for federal security review up to 90 days before public release. It was the first concrete attempt by this administration to formalise pre-release vetting of frontier systems, and it borrowed structurally from a Biden-era voluntary commitments framework that the White House had previously dismantled.

What the industry told the president

According to CNN, calls from Mark Zuckerberg, Elon Musk, and former White House AI and crypto adviser David Sacks reached the president in the hours before the signing. The argument they pressed was familiar. Any pre-release review window, even a voluntary one, risks slowing US development at a moment when Beijing is closing the gap on frontier capabilities.

Trump told reporters afterwards that the order "gets in the way," adding that "the United States is leading China, we're leading everybody." He said he had postponed rather than killed the order because there were "certain aspects" he wanted to revisit.

What the order would have done

The text, summarised in The Washington Times, established a framework under which the federal government could request pre-release access to test high-capability models for cybersecurity, biosecurity and dual-use risks. Compliance was voluntary. The Center for AI Standards and Innovation, the safety-focused successor to the AI Safety Institute, would have run the evaluations.

Even framed as voluntary, the proposal was a meaningful shift in posture from an administration that has spent its first months stripping back the Biden-era safety scaffolding around AI. The reversal hours before signing leaves the field without any formal mechanism for federal review of frontier models.

The wider regulatory vacuum

The postponement deepens an uneven US regulatory picture in which states are now setting most of the substantive rules. Connecticut has passed one of the country's most comprehensive AI bills, Iowa has enacted a chatbot safety law, and Colorado has bills moving on chatbot safety and dynamic pricing. A recent Fortune analysis counted more than 1,200 AI-related state bills in active circulation.

For the frontier labs themselves, the immediate effect is that nothing changes. There is no mandatory pre-release testing, no federal evaluator, and no legal floor on what a developer must disclose before shipping a new model.