White House seeks early model access

President Donald Trump’s administration is preparing an executive order on AI and cybersecurity that would give the federal government earlier visibility into advanced AI systems, according to Axios. The order could be released as soon as this week, Axios reported on May 20, citing sources familiar with the matter. Later coverage from Gizmodo said the latest draft would make model-sharing with the government voluntary rather than mandatory. The emerging language follows weeks of reporting that White House officials were weighing a tougher pre-release review regime for frontier models. ### What is the White House trying to get from AI companies? Axios reported on May 20 that the draft order would create a voluntary framework for AI developers to notify the government about new frontier-model releases. The same report said an earlier version had gone further by seeking government access to covered models before public deployment. (axios.com) Gizmodo reported later on May 20 that the final order was expected to soften that approach by making sharing voluntary. Its report described the change as a retreat from broader earlier language that had raised the prospect of stronger federal access demands. ### How does this compare with what the administration was considering earlier this month? (axios.com) Politico reported on May 5 that the Trump administration was discussing tighter controls on advanced AI, including a vetting regime focused on national security risks. CNBC reported the same day that the White House was considering an AI working group that could explore procedures for reviewing models before release. (gizmodo.com) Bloomberg reported on May 8 that the administration was preparing an AI security order that would stop short of requiring government approval for cutting-edge models. That reporting pointed to a narrower approach centered on partnerships between agencies and AI companies rather than a formal licensing system. (politico.com) ### Which agencies are already involved in pre-release model testing? The Commerce Department’s AI safety apparatus was already moving in that direction before the draft order surfaced. Forbes reported on May 5 that Google, Microsoft and xAI had reached agreements allowing the Trump administration to review new AI models before release. CNBC separately reported that the White House was weighing a broader working group on oversight alongside those arrangements. (bloomberg.com) Webhound, summarizing a May 5 NIST announcement, said the government action published that week involved voluntary agreements with Google DeepMind, Microsoft and xAI for pre-deployment evaluations and related research. That arrangement involved the Center for AI Standards and Innovation, or CAISI. ### Why is the order tied to cybersecurity? (forbes.com) Axios reported that the executive order is framed around cybersecurity and AI safety and is intended to tighten security around advanced models while improving threat-sharing between industry and government. Anadolu, citing Axios, said the draft contemplated developers sharing new models with the government at least 90 days before public release under a voluntary framework. (webhound.ai) Federal News Network reported on May 6 that Kevin Hassett, director of the National Economic Council, said the administration was considering an order to ensure new AI models are secure before public release. Hassett compared the approach to how the Food and Drug Administration evaluates drugs for safety, according to that report. (axios.com) ### What happens next? Axios reported that the White House could release the order as soon as the week of May 20. The final text will determine whether developers are asked to participate through a voluntary disclosure system, a structured review process, or narrower cybersecurity partnerships already under discussion. (axios.com) (federalnewsnetwork.com)