Unlock the White House Watch newsletter for free
Your guide to what Trump’s second term means for Washington, business and the world
The White House has unexpectedly postponed the signing of a long-awaited executive order on AI, after Donald Trump said he “did not like” aspects of the plan for the US to vet models for national security and cyber risks.
The order, which would have seen leading AI companies such as OpenAI, Google and Anthropic voluntarily commit to submit their models for government checks, was due to be signed on Thursday afternoon.
The president’s U-turn came after weeks of infighting within the administration over the ideal scope of regulatory oversight.
“I didn’t like certain aspects of it,” Trump said of the order. “We’re leading China, we’re leading everybody, and I don’t want anything that’s going to get in the way of that lead.”
AI was also “bringing in a lot of jobs”, he said.
Several tech chief executives were due to join Trump in Washington for the signing before it was abruptly postponed.
The postponement comes as polls consistently show American voters are concerned about the effects of AI, and that many support tougher regulations and guardrails for the nascent technology.
Public concern about the safety implications of unleashing powerful AI models — alongside worries about job losses and vast data centre construction — has complicated the political calculus for an administration that had until recently been enthusiastically pro-AI.
Some of the president’s allies have called for leading models to be brought under US government control, while other Maga figures have warned that measures that limit the growth of AI would hinder the economy.
A poll carried out this month for the Institute for Family Studies found 82 per cent of Americans backed White House safety testing for advanced models.
The order was set in train by the White House after key officials including Treasury secretary Scott Bessent were given an early look at latest Anthropic’s Mythos model, which has advanced capabilities to detect cyber security weaknesses. Officials were alarmed by the vulnerabilities it exposed in the banking system among other issues.
Anthropic has so far limited the release of Mythos to a small number of trusted institutions, including tech companies and some banks, so they can detect and fix cyber issues before hackers can gain access to the model.
At one point, Kevin Hassett, Trump’s director of the National Economic Council, suggested that frontier models ought to be subject to the same kind of approval process as pharmaceuticals, and released only “after they’ve been proven safe, just like an FDA drug”.
His comments were met with fierce opposition from AI founders and investors, including some close to the Trump administration, who argued such a regime would hamper US innovation.
The executive order stopped far short of such approval requirements, and instead set up “a collaborative, voluntary framework for benchmarking models”. Officials briefed journalists on the contents of the order on Thursday morning in anticipation of the planned signing.
Under the proposed agreement, leading AI companies including OpenAI and xAI would voluntarily share their models with the government 90 days before they are due to be rolled out to the public, with the administration ultimately relying on the goodwill of AI bosses.
Credit: Source link
