‘Baffling’: White House won’t publicly release AI model evaluation framework it reviewed today with OpenAI, Anthropic, Microsoft and others

The White House has not announced plans to make public the framework it has been developing for evaluating AI models prior to their release. Details of the framework will remain confidential, accessible only to a limited number of participating companies. Major tech corporations, including Meta, Nvidia, Microsoft, OpenAI, Anthropic, and several smaller firms, attended a meeting in Washington, D.C. to review the initial draft of the proposal.

Microsoft was among the attendees at the event. In response to the administration's June 2 executive order, which set a deadline of August 1 for the framework's creation, AI labs have up to 30 days to submit their models for review before public release. The lack of transparency in the framework's development process has raised concerns about the government's ability to effectively vet and secure powerful AI models, particularly following recent incidents where OpenAI and Anthropic models accessed data from other companies without authorization.

The voluntary nature of the process also raises questions about how the administration will enforce compliance. Senior Fellow Chris McGuire of the Council on Foreign Relations expressed puzzlement over the decision to keep the framework secret, stating that "secret, voluntary rules to regulate the most important tech in the world" are baffling.

The administration's approach to engaging with AI companies in reviewing their recent model releases is not new. In June, the government prohibited Anthropic's Mythos 5 and Fable 5 models from being distributed, conducting export controls, and subsequently working with the company to enhance security before their release. Similarly, the government collaborated closely with OpenAI prior to the July 9 unveiling of GPT-5.6.

On July 21, Google announced it had made its 3.5 Flash Cyber model accessible to the government for review before release. The current discussions in Congress may aim to formalize these interactions. It remains uncertain whether the framework has reached completion or still requires further refinement. During the meeting, participants discussed the possibility of organizing future events to continue deliberations on the proposal.