Trump weighs pre-release AI review

President Donald Trump is considering a voluntary process that would let the U.S. government review powerful artificial intelligence models before they are released to the public, according to multiple reports this week. The proposed framework is aimed at national-security and cybersecurity risks tied to so-called frontier models, the most advanced systems being developed by companies including OpenAI and Anthropic. Reuters reported on May 20 that Trump was expected to sign a broader executive order on AI and cybersecurity as soon as Thursday, though the order’s final form was still being worked out. The White House was also trying to assemble AI company chief executives for a signing ceremony, Reuters said. ### What exactly is the administration considering before model launches? The White House is weighing a voluntary government review system for advanced AI models before public release, OpenTheMagazine reported on May 21, citing CNN. Under that approach, companies would give U.S. authorities access to frontier systems ahead of launch so officials could assess national-security and cybersecurity risks. Politico reported that a draft directive would ask technology companies to submit advanced models for review by federal agencies, while Bloomberg reported earlier this month that the administration was preparing an AI security order that would stop short of requiring government approval for cutting-edge models. (usnews.com) A Reuters report published May 20 said the broader executive order would address AI and cybersecurity as pressure grew from parts of Trump’s political base for stronger oversight of new models. Reuters said those concerns had centered in part on systems such as Anthropic’s Mythos. ### Why is the review described as voluntary, not mandatory? (openthemagazine.com) Bloomberg reported on May 8 that the administration’s planned directive would omit mandatory model tests and instead focus on partnerships between federal agencies and AI companies to protect networks from AI-enabled cyberattacks. That aligns with later reports describing a disclosure or review regime that companies would join voluntarily rather than through a formal licensing requirement. (usnews.com) The distinction matters because several reports said administration officials had debated tougher options. Politico reported on May 5 that the White House had discussed a vetting regime for frontier models on national-security grounds, and CNBC reported the administration was considering a working group that could explore oversight procedures before release. (bloomberg.com) ### Who is pushing for tighter scrutiny of frontier AI? Reuters said on May 20 that pressure for stronger oversight was coming from parts of Trump’s political base. Gizmodo reported that Steve Bannon and more than 60 Trump allies had urged the president to require federal vetting of what they called dangerous frontier AI models before release. Reuters did not independently confirm that letter in the search results reviewed here, but its reporting and other outlets’ accounts point to a coordinated push from Trump-aligned figures for tighter screening. (politico.com) Politico also reported that the administration’s draft order would increase federal involvement in reviewing advanced systems. That would put Washington closer to the center of launch decisions for the most capable models, even if the program remains formally voluntary. ### Where does OpenAI fit into this story? OpenAI is expanding its legal bench as the regulatory and litigation stakes rise. (usnews.com) Reuters reported on May 21 that the company had added to its roster of law firms to handle major lawsuits and deals after a court victory for OpenAI, Chief Executive Sam Altman and their lawyers in Elon Musk’s case. Reuters said the ruling removed a potential obstacle to an OpenAI initial public offering. (politico.com) Bloomberg reported on May 20 that OpenAI was aiming to file for an IPO in the coming days or weeks, with a goal of going public in the fall, according to a person familiar with the matter. That timetable has not been confirmed by OpenAI in the sources reviewed here. ### What should readers watch next? Thursday is the immediate date to watch because Reuters said Trump could sign an executive order on AI and cybersecurity as soon as then, depending on final White House planning. (economictimes.indiatimes.com) The next concrete milestone after that will be the text of any order, which would show whether the administration settles on voluntary pre-release reviews, broader cyber provisions, or both. OpenAI’s next step is separate but related: Reuters said its legal expansion follows a court win, while Bloomberg reported an IPO filing could come in the next several days or weeks. (bloomberg.com) (usnews.com)