White House delays AI order

President Donald Trump said on Thursday, May 21, that he had postponed signing a planned executive order on artificial intelligence after deciding he did not like parts of the draft and did not want to risk slowing the United States in its competition with China. The order had been expected to create a voluntary framework for AI developers to engage with the U.S. government before releasing advanced models, according to Reuters and other reports. Trump made the decision hours before a planned White House signing ceremony that had been expected to include chief executives from AI companies. (usnews.com) As of Friday, May 22, the White House had not posted any AI executive order on its public executive-orders page. ### What did Trump say he objected to? Trump told reporters in the Oval Office that he delayed the order because “I didn’t like certain aspects” of it and because he did not want to do anything that would get in the way of the U.S. lead in AI over China. Reuters reported that Trump did not specify which provisions he opposed. The South China Morning Post reported that the administration had been preparing a broader regulatory step for the fast-growing AI market, but that concern about U.S. competitiveness with global rivals led Trump to pull back. (usnews.com) Reuters reported the same concern in narrower terms, saying Trump did not want to undermine the U.S. position in its AI competition with China. ### What was the order expected to do? Reuters reported that the draft order would create a voluntary framework for AI developers to engage with the U.S. government before the public release of advanced AI models. (usnews.com) The South China Morning Post said earlier reports described a requirement for developers to submit models to several federal agencies for voluntary review up to 90 days before public release. Reuters also reported that Trump had planned to direct the U.S. government to use advanced AI models to improve cybersecurity defenses for federal systems and for networks in sectors such as banks and hospitals. (scmp.com) That part of the proposal reflected growing concern inside government and industry about the security risks posed by more capable models. ### Why were security concerns part of the debate? Anthropic’s model Mythos was cited in both Reuters and the South China Morning Post as a recent example of why officials had been weighing stronger oversight. (usnews.com) Reuters reported that concerns were growing across the U.S. government and private sector about cybersecurity risks from powerful new AI systems, and said Anthropic had warned Mythos could intensify complex cyberattacks. The South China Morning Post reported that Anthropic did not publicly release Mythos and instead provided it to a consortium of U.S. companies including Cisco, JPMorgan Chase and Nvidia through a program called Project Glasswing aimed at protecting critical software. (usnews.com) ### What evidence is there that current systems remain vulnerable? A BBC-linked investigation published in February 2025 found that major chatbots struggled to summarize news accurately and that more than half of tested answers contained significant issues, with factual errors introduced in about a fifth of answers. (usnews.com) The testing covered ChatGPT, Microsoft Copilot, Google Gemini and Perplexity, according to reports describing the BBC study. (scmp.com) Those findings do not describe the delayed White House order directly, but they show the kind of reliability and misinformation problems that have continued to shape the policy debate around advanced AI systems. Deborah Turness, chief executive of BBC News and Current Affairs, said at the time that tech firms were “playing with fire” if inaccurate AI-generated news answers were allowed to persist, according to reports on the study. (business-humanrights.org) ### How does this fit with Trump’s broader AI approach? Reuters reported that Trump has taken a softer stance toward Big Tech companies than former President Joe Biden, even as some prominent Trump supporters have called for guardrails around AI. The White House had been expected to unveil a framework that mixed voluntary review with cybersecurity uses for advanced models, but the postponement left that balance unresolved for now. A separate South China Morning Post report from March said the administration had already been pursuing an AI policy framework aimed at limiting state-level restrictions on innovation while urging Congress to address risks and preserve the U.S. edge over China. (business-humanrights.org) That earlier report suggests the administration’s internal debate has centered less on whether to act than on how much oversight to impose without slowing deployment. ### What happens next? (usnews.com) May 22 is the clearest public marker so far: the White House executive-orders page still does not list the delayed AI action. Reuters reported on May 21 that Trump had planned to sign the order that afternoon, but no new date was announced in the reporting reviewed here. Any next step is likely to come either through a rescheduled White House signing or publication of a revised order by the administration. (whitehouse.gov) (scmp.com)