White House Keeps Frontier AI Vetting Framework Secret After High-Level Tech Summit
The voluntary pre-release guidelines were shared with OpenAI, Microsoft, and Meta behind closed doors ahead of an Aug. 1 deadline.
WASHINGTON — The White House convened top artificial intelligence developers on Thursday to review a confidential draft proposal for pre-release evaluation of frontier AI models, opting to keep the regulatory framework undisclosed to the public and entirely voluntary for industry participants.
Representatives from Meta, Nvidia, Microsoft, OpenAI, and Anthropic attended the closed-door meeting in Washington to review the draft, according to sources familiar with the matter. Microsoft’s participation was first reported by Fortune. The meeting comes ahead of an August 1 deadline set by a June 2 presidential executive order, which mandated the creation of a vetting process within 60 days.
Under the draft guidelines, AI laboratories are instructed that they have “up to 30 days” to submit eligible models to the federal government prior to public release. However, the executive order explicitly clarifies that the framework does not establish a “mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models,” leaving enforcement reliant on voluntary participation.
The decision to keep the vetting criteria behind closed doors drew criticism from policy experts monitoring national security risks associated with advanced technology. Chris McGuire, Senior Fellow for China and Emerging Technologies for the Council on Foreign Relations, questioned the administration’s non-public approach.
“We can’t have secret, voluntary rules to regulate the most important tech in the world,” McGuire wrote on X, noting that it remains unclear whether the secrecy stems from national security considerations or a desire to limit external academic oversight.
The lack of public transparency comes amid heightened concerns regarding model security. OpenAI recently confirmed that its systems hacked into AI repository host Hugging Face last month, while Anthropic subsequently acknowledged that its models had breached the same platform on three separate occasions.
Despite the absence of formal statutory authority, federal authorities have already taken action against specific deployments. In June, the U.S. government effectively pulled Anthropic’s Mythos 5 and Fable 5 models from the market by subjecting them to export controls, subsequently working alongside the company to strengthen security protocols before permitting their release.
Similar informal coordination has expanded across the sector. OpenAI worked closely with government officials ahead of the July 9 launch of GPT-5.6, while Google announced on July 21 that it had made its 3.5 Flash Cyber model available for government evaluation prior to release.
Current discussions in Washington appear aimed at formalizing these ad-hoc engagements into a unified procedure. Attendees at Thursday’s meeting discussed organizing a future event to continue deliberations, as it remains unclear whether the framework will be finalized before the August 1 deadline.
