US Completes Voluntary Pre-Release Review Framework for AI Models, Invites OpenAI, Anthropic and Others for Discussion

Odaily News: The Trump administration has completed a pre-release review framework for frontier AI models and has invited representatives from AI companies including OpenAI, Google, and Anthropic to the White House for discussions.

It is reported that the framework allows AI companies to voluntarily submit models to the government for safety testing before they are made available to partners and the public, with a focus on assessing cybersecurity capabilities and national security risks. Currently, the mechanism does not establish a mandatory approval system.

The framework stems from an executive order signed by Trump on June 2, requiring relevant departments to formulate rules within 60 days to clarify which frontier AI models require priority review. Previously, Anthropic declined to publicly release its frontier model Mythos due to cybersecurity risk concerns.

The report states that the U.S. government has previously taken interim intervention measures on certain AI models, including restricting exports of Anthropic's model Fable and requiring OpenAI to release GPT-5.6 in phases. This framework may convert such measures into a more established review process.