Trump preparing AI executive order

President Donald Trump is preparing an executive order that would tighten federal oversight of advanced artificial intelligence models and add new cybersecurity-focused scrutiny before some systems are released. Reuters reported on May 21 that the White House could roll out the order this week after internal debate among Trump allies, tech executives and security officials over how much precaution to impose on frontier AI systems. The draft effort would mark a more interventionist step than Trump’s January 2025 AI order, which emphasized removing barriers to innovation. ### Which AI systems pushed the White House toward a new order? OpenAI’s GPT-5.5-Cyber and Anthropic’s Mythos were cited by officials in the Reuters report as examples of models that raised alarm inside the administration. The concern, according to Reuters, is that those systems could materially improve the sophistication of cyberattacks by helping users identify vulnerabilities, analyze malware or automate parts of offensive cyber work. Politico reported on May 7 that OpenAI had begun rolling out GPT-5.5-Cyber to vetted cybersecurity professionals, positioning it against Anthropic’s Mythos. OpenAI said the model was designed for defensive uses such as finding and patching vulnerabilities, but the administration’s concern is that the same capabilities could be misused if access broadens or safeguards fail. ### What would the executive order actually require? CNN, cited in multiple follow-up reports on May 21, said the draft order would create a voluntary process for companies to notify the government before releasing powerful new AI models. Politico separately reported on May 5 that the administration had been discussing a vetting regime for frontier systems to examine national security risks. Gizmodo reported on May 21 that one draft version would make model sharing with the government voluntary rather than mandatory. Reuters said the package under discussion centers on AI oversight and cybersecurity for advanced models, though the final language and enforcement structure had not been released as of Thursday. ### Why is this a change from Trump’s earlier AI posture? Trump signed an executive order on January 23, 2025, aimed at promoting AI systems “free from ideological bias” and revoking prior policies his administration said hindered innovation. That order directed the development of an AI action plan within 180 days and framed federal policy around U.S. leadership and deregulation. The newer proposal is narrower and more security-focused. Reuters said the shift followed warnings from companies including OpenAI and Anthropic, along with pressure from some Trump supporters who wanted the administration to take frontier-model risks more seriously. ### What is OpenAI doing at the same time in the states? Chris Lehane, OpenAI’s chief global affairs officer, is urging Democratic-led states to pass laws that align with the company’s preferred national framework, Politico reported on May 20. Politico said Lehane has argued that state legislation can help build momentum for a broader federal approach while Congress remains stalled. That state-level push comes after a longer campaign by major AI companies to shape Washington’s response to state regulation. Politico reported in 2025 that OpenAI, Meta, Google and IBM had all backed federal rules that could limit or override some state AI laws they opposed. ### What happens next? The White House could issue the executive order as soon as Thursday, according to CNN and Reuters-based reports published on May 21. If Trump signs it, the next test will be whether companies such as OpenAI and Anthropic participate in any voluntary pre-release review process and how federal agencies define which models qualify as advanced enough to trigger scrutiny.