The Trump administration is considering a new executive order that would establish government oversight for advanced artificial intelligence (AI) models before they reach the public in a radical shift from its staunch deregulatory stance.
According to a The New York Times report, White House officials are in the early stages of deliberating a federal review process. The proposed plan involves the creation of an AI working group of government officials and tech industry executives to develop standardized oversight procedures.
The discussions represent a notable pivot for the administration. Immediately upon taking office, President Trump revoked a Biden-era executive order aimed at managing AI risks, favoring a more hands-off approach. However, the current deliberations suggest a growing concern regarding the national security implications of frontier models.
The change in policy follows a change in leadership within the White House’s tech policy sphere. Following the March departure of David Sacks, the administration’s AI Czar who championed deregulation, Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent have reportedly taken more active roles in shaping the country’s AI strategy.
Sources indicate the proposed framework may mirror the United Kingdom’s AI Safety Institute. Under this structure, government bodies — including the NSA, the Office of the National Cyber Director, and the Director of National Intelligence — would evaluate models against specific safety benchmarks.
Under the proposed plan, the government would gain access to new models during the development phase; models would be tested for risks related to cybersecurity and national security; and aim to vet models without necessarily blocking their public release.
“This would be a huge blow to companies used to launching quickly and competing head-to-head on model drops, benchmark wins, and feature velocity,” Kate Shen, co-founder of Anaxi Labs, said in an email. “A mandatory review layer would significantly slow releases, raise compliance costs, leak strategic timing advantages, and favor incumbents that can afford policy, legal, red-team, and government-relations infrastructure. It will also change product culture: ‘ship fast, patch later’ becomes ‘prove safety, then ship.’”
The catalyst for this sudden interest in oversight appears to be the release of Anthropic’s Mythos model. Anthropic’s marketing of the tool, claiming it could identify thousands of critical software vulnerabilities, branded the technology as “too dangerous” for unrestricted public use.
This framing caught the attention of federal agencies, particularly as the administration remains in a legal and contractual deadlock with Anthropic. The Pentagon recently designated the startup a “supply chain risk” after Anthropic refused to remove ethical guardrails regarding autonomous weaponry, a move a federal judge later described as Orwellian.
Despite the friction, the NSA has already reportedly used Mythos to scan government software for vulnerabilities, highlighting the dual-use nature of the technology that is driving the White House toward a formal review process.
While the administration has briefed leaders at OpenAI, Google, and Anthropic on these potential changes, the White House has publicly characterized the reports as speculation, noting that any formal policy shift would come directly from the president.
Dean Ball, a former senior AI adviser to the administration, noted that officials are currently navigating a “tricky balance” between maintaining American technological dominance and ensuring that rapidly evolving AI does not compromise national security.
“The goal should be regulation that acts like an air-traffic-control system: not stopping flights, but making high-speed movement safer, predictable, and globally interoperable,” Shen said.
