The draft executive order, expected as early as Thursday, introduces a voluntary 90-day pre-release model-disclosure framework with federal authorities, involving early participation by critical-infrastructure providers like banks. Steve Bannon and Amy Kremer have advocated for a stricter, mandatory approach.
President Donald Trump is poised to sign an executive order on AI oversight as soon as Thursday, Reuters reported on Wednesday, amid increasing demands from certain political factions seeking more stringent security reviews of advanced AI systems.
The order outlines a voluntary framework where AI developers would grant the federal government pre-release access to advanced models 90 days prior to public release, with critical-infrastructure operators such as banks also receiving early access.
The political momentum behind the order’s progression is notable for its direction. U.S. News & World Report’s insight into the order highlights a coalition within the broader MAGA constituency, including ex-Trump adviser Steve Bannon and right-wing political organizer Amy Kremer, who have been urging the White House to enforce mandatory security checks for the most capable models.
Bannon frames the issue, noting that the launches of Anthropic’s Mythos and OpenAI’s GPT-5.5-Cyber have altered the cybersecurity landscape in a way that requires federal attention.
The 💜 of EU tech
Stay updated with the latest EU tech news, a story from our insightful founder Boris, and some intriguing AI art. Delivered weekly to your inbox, free of charge. Sign up now!
The order’s voluntary nature reflects a compromise within the West Wing. Advisers focused on industry concerns, such as David Sacks, opposed mandatory disclosure on competitiveness grounds, while Bannon-Kremer-aligned national security figures pushed for more stringent measures.
The resulting voluntary 90-day framework is a product of these discussions. The draft does not impose penalties for opting out, although non-participation could be publicly noted by the government.
The model-disclosure framework addresses a particular cybersecurity threat landscape. Anthropic’s previous commitment to share Mythos findings with partner governments, including recent FSB-side briefings on Mythos cyber capabilities, established a precedent now formalized in the U.S. context by the order.
The Trump administration maintains that the voluntary approach will provide federal cybersecurity entities (CISA, NSA’s Cybersecurity Collaboration Center, and parts of the FBI) with a pre-release window similar to that already provided to foreign intelligence services by the Mythos initiative.
Early access for critical infrastructure is a segment of the order expected to be swiftly implemented. Vorys’s legal analysis of the White House AI governance strategy indicates that critical infrastructure encompasses banks, energy operators, telecom providers, and major healthcare networks.
The 90-day period aims to allow these operators to test and secure their defenses against newly developed models before they become widely available.
The order comes amid heightened political