Trump proposes 90-day model sharing

President Donald Trump is expected to sign an executive order on Thursday, May 21, aimed at artificial intelligence cybersecurity, according to Bloomberg and Axios. The reporting points to a federal push to bring AI companies into existing cybersecurity information-sharing channels, not a formal pre-approval regime for advanced models. That distinction matters because social media posts this week described a much broader plan requiring frontier labs to hand models to the government 90 days before release. As of Thursday morning, the White House executive orders page did not show a new AI cybersecurity order. ### Where did the “90 days before release” claim come from? Alex Heath, a deputy editor at The Verge, posted on X that Trump planned an AI safety executive order requiring frontier labs to share models with the government 90 days before release and listed invited executives including Elon Musk, Sam Altman, Dario Amodei, Mark Zuckerberg, Sundar Pichai and Tim Cook, according to the social briefing provided for this story. The same social chatter referenced model names including “Mythos” and “GPT-5.5-Cyber,” and said U.S. (bloomberg.com) Cyber Command and the National Security Agency would launch an AI task force. The social posts are the origin of the most sweeping claims now circulating. ### What have reported outlets actually confirmed? Bloomberg reported on May 20 that Trump was poised to issue an executive order “as soon as Thursday” to bolster AI cybersecurity and had asked tech industry leaders to join the event, citing people familiar with the matter. Bloomberg also reported earlier, on May 8, that the draft would revamp existing cybersecurity information-sharing programs to include AI companies and would stop short of requiring government approval for cutting-edge models. (axios.com) Axios reported on May 19 that the current draft outlined a voluntary framework for AI developers to inform the government about new releases. ### Is there evidence of a mandatory model handover rule? Axios described the framework as voluntary, and Bloomberg reported that the order would stop short of mandatory federal approval of advanced models. Neither outlet, in the material available here, confirmed a requirement for labs to turn over models 90 days before release. That means the strongest independently reported version of the policy is narrower than the social-media version. (bloomberg.com) ### Why are “Mythos” and cyber risk showing up in the reporting? Bloomberg reported on May 8 that the directive took shape about a month after Anthropic revealed its Mythos model. In that same report, Bloomberg said the administration was preparing to have U.S. agencies partner with AI companies to protect networks from AI-enabled cyber attacks. That places cyber misuse risk — rather than broad consumer AI regulation — at the center of the order as reported by mainstream outlets. (axios.com) ### How does this fit Trump’s broader AI policy? The White House has already issued AI-related actions during Trump’s second term, including a January 2025 order on “Removing Barriers to American Leadership in Artificial Intelligence,” a July 2025 AI Action Plan release, and a December 2025 order on a national policy framework for AI. Those documents emphasize U.S. leadership, federal coordination and limits on conflicting state rules. The reported cybersecurity order would add a national-security layer focused on advanced models and cyber threats. (bloomberg.com) ### What should readers watch for next? May 21 is the key date. The cleanest confirmation will be a White House executive order, fact sheet, or event readout naming the participating companies and spelling out whether reporting to the government is voluntary or mandatory. Until that appears, the best-supported version is this: Trump is preparing an AI cybersecurity directive, but the reported text available from Bloomberg and Axios does not confirm the full 90-day mandatory model-sharing claim circulating on social media. (whitehouse.gov) (bloomberg.com)