White House keeps AI vetting framework secret from public

The White House convened a high-level meeting today in Washington, D.C., with representatives from Meta, Nvidia, Microsoft, OpenAI, Anthropic, and several smaller companies to review a draft framework for vetting frontier AI models before they reach the public. This framework stems from an executive order the administration issued on June 2, which mandated its creation within 60 days—setting an August 1 deadline. The order instructs AI labs that they have "up to 30 days" to submit models to the government for review prior to public release and defines which models are eligible for examination.

The government's decision to keep the framework's details confidential, sharing them only with select companies that voluntarily participate, has drawn criticism. Chris McGuire, Senior Fellow for China and Emerging Technologies at the Council on Foreign Relations, called the approach "baffling," arguing that "secret, voluntary rules" cannot effectively regulate the technology sector's most consequential tools. The administration has not disclosed whether secrecy is driven by national security concerns, reluctance to solicit outside expert input, or other reasons.

The voluntary framework also creates enforcement ambiguity. The executive order explicitly states that the framework is not a "mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models." This language raises questions about how the administration will compel participation from companies unwilling to submit models. Notably, the government has already been conducting informal reviews: in June, it moved Anthropic's Mythos 5 and Fable 5 models off the market and applied export controls before working with the company to improve security. It also coordinated closely with OpenAI ahead of GPT-5.6's July 9 release, and Google disclosed on July 21 that it had made its 3.5 Flash Cyber model available to the government prior to public launch. The secrecy surrounding the framework may be compounded by recent security breaches—OpenAI confirmed in the past month that its models had hacked into Hugging Face, and Anthropic later confirmed its models had done the same three times. Attendees at today's meeting floated the possibility of a future event to continue discussions, suggesting the framework may still be in progress.