White House Reviews Pre-Release Screening of AI Models

The White House is reviewing the introduction of a government screening process before AI models are released.

According to U.S. officials and people familiar with the matter, the administration is discussing setting up an AI working group involving both the tech industry and government, along with formal review of new models. Recently, the White House briefed executives at Anthropic, Google, OpenAI on the related initiative.

• The review's focus is on models' cybersecurity capabilities and potential use by defense and intelligence agencies.
• Some proposals would have the government check models first while not blocking the release itself.
• Candidate agencies mentioned include NSA, the Office of the National Cyber Director, and DNI.
• The role of the Center for AI Standards and Innovation, created under the Biden era, is also being reconsidered.

Behind this is the fact that Anthropic's new model Mythos showed cyber capabilities strong enough to delay its release, and the White House's judgment aiming to avoid the political fallout of a large-scale AI-driven cyberattack. However, this is not yet a finalized policy, and the White House drew a line stating that discussion of a related executive order is speculation. This discussion is being read as a clear pivot from the Trump administration's existing stance of deregulation.