Trump prepares sweeping AI order

President Donald Trump was expected to sign an executive order on artificial intelligence and cybersecurity as soon as Thursday, May 21, according to Reuters, as White House officials weighed a new federal process for reviewing advanced AI systems before release. The draft plan under discussion would create a government framework for companies developing the most powerful models to engage with federal officials ahead of deployment. Reuters reported the approach had emerged after weeks of internal White House work and pressure from rival factions around Trump over how hard the administration should press leading AI firms. The debate has centered on companies including OpenAI and Anthropic, whose newest systems have sharpened concerns in Washington about cyber and national security risks. ### What would the order require from AI companies? Reuters reported the order would set up a voluntary framework under which developers of “covered” models would be asked to provide those systems to the U.S. government 90 days before public release. One source told Reuters the same framework would also give critical infrastructure providers such as banks pre-public access to those models. (usnews.com) Politico reported earlier this month that the White House had discussed a more stringent vetting regime that could require companies to receive a government green light before releasing advanced models. National Economic Council Director Kevin Hassett told Fox Business the administration was studying an order that would create a process so future AI systems would be released only after they had been “proven safe,” comparing the concept to drug review. (usnews.com) ### Why is the administration acting now? May 20 Reuters reporting said pressure had grown inside Trump’s political coalition to tighten oversight of powerful new systems, including Anthropic’s Mythos. Reuters said former Trump adviser Steve Bannon and organizer Amy Kremer were among those pressing for government security tests, while venture capitalist Marc Andreessen and former Trump adviser David Sacks opposed mandatory requirements. (politico.com) May 5 reporting from Politico said the release of newer frontier systems had accelerated White House deliberations. The outlet reported the administration had already signed agreements with Microsoft, xAI and Google DeepMind allowing the government to evaluate their models for national security risks ahead of release. (usnews.com) ### Where do OpenAI and Anthropic fit into this? OpenAI and Anthropic sit near the center of the debate because both companies are releasing increasingly capable models and both have been drawn into Washington’s argument over how much oversight frontier AI should face. Reuters said the new order was being developed as pressure mounted to increase scrutiny of models such as Mythos and OpenAI’s GPT-5.5-Cyber. (politico.com) Politico reported on May 20 that OpenAI’s chief global affairs officer, Chris Lehane, is now pursuing what he called “reverse federalism” — a state-by-state lobbying strategy meant to shape AI laws in large states and build a de facto national standard. Lehane told Politico OpenAI had already found success in California and New York and was turning next to Illinois. Politico said the company-backed approach emphasizes transparency and reporting requirements for advanced-model developers while imposing fewer new liabilities than some AI safety advocates want. (usnews.com) ### Has the White House changed its approach? Bloomberg reported on May 8 that an earlier draft of the administration’s AI security order stopped short of requiring government approval for cutting-edge models and focused more heavily on partnerships between agencies and AI companies to protect networks from AI-enabled cyberattacks. Reuters’ May 20 account described a framework that still appears voluntary but would ask companies to share models before release, suggesting the White House was trying to bridge competing demands inside Trump’s orbit. (politico.com) That is an inference based on the two reports’ descriptions of the draft and the later proposal. Reuters said the order was shaped over the last month by White House Chief of Staff Susie Wiles, White House science and technology adviser Michael Kratsios, Wiles deputy Walker Barrett and National Cyber Director Sean Cairncross, with input from AI companies. A White House spokesperson called discussion of policy details “speculation.” (bloomberg.com) ### What happens next? Thursday, May 21, was the earliest expected signing date cited by Reuters, and the White House was working to bring AI company chief executives to a ceremony with Trump. The next concrete test will be the text of the order itself: whether it keeps a voluntary review system, adopts tougher testing requirements, or adds broader national security controls for companies including OpenAI and Anthropic. (usnews.com)