White House Reviews Pre-Release Screening of AI Models

The White House is discussing an executive order that would establish a pre-screening system where the government reviews AI models before public release. The plan under discussion involves creating an AI Working Group to verify the safety and cyber capabilities of new models. It is also referencing the UK's approach, where multiple government agencies review safety standards.

The White House has already shared this concept with executives at Anthropic, Google, and OpenAI. This marks a notable shift from the deregulatory stance the Trump administration has maintained since taking office, and industry insiders worry that excessive intervention could slow the pace of U.S. innovation in competition with China. A Pew Research Center survey also found that 50% of Republicans and 51% of Democrats said they are more concerned about the spread of AI.

The trigger for this discussion was Anthropic's new model, Claude Mythos Preview. Its vulnerability detection capabilities were so strong that its release was delayed, and the White House is reportedly concerned about the political fallout should a major AI-driven cyberattack occur. Some officials are considering an approach where the government would receive models first without necessarily blocking their release.

The axis of coordination within the administration is also shifting. Since David Sacks stepped down in March, Susie Wiles and Scott Bessent have been more directly overseeing policy. Meanwhile, a conflict over a $200 million contract between the Department of Defense and Anthropic led to a suspension of government use in March, though the technology continues to be used in the military system Maven and for vulnerability checks at the NSA. Candidate bodies being discussed include the NSA, the White House Office of the National Cyber Director, and the Director of National Intelligence (DNI), with the role of the Center for AI Standards and Innovation also back under discussion.