According to sources, the White House is not planning to publicly release its newly developed advanced AI model evaluation framework. This voluntary framework has significant implications for global AI safety, but specific details will only be shared with relevant companies involved in the development process.

This means that companies, policymakers, researchers, and U.S. overseas allies outside of this process will not have an accurate understanding of how the White House will implement its key AI policies. On Tuesday, the White House held a working-level meeting with industry representatives to communicate about the framework, but companies not invited to the meeting remain unaware of the specific content of the framework.

Executive Order First, Security Evaluation Standards Have Been Classified as Confidential

The release of this framework comes amid the AI industry's deepening challenges, including frequent high-risk cyberattacks and the rapid technological advancements by Chinese AI developers. According to an executive order issued in June, this voluntary framework aims to provide AI developers with a mechanism to collaborate with the government to determine whether their models fall under regulatory scope.

The framework is also expected to establish rules allowing the government to access relevant information within up to 30 days before the release of an AI model, covering requirements such as confidentiality, cybersecurity, internal personnel risks, intellectual property, and non-disclosure agreements. The executive order explicitly states that the technical benchmark testing process used to evaluate AI model network capabilities will be classified as confidential, and the order itself does not require the voluntary framework to be made public.

Open Source Issues Rise Again, Huang Renxun Visits Washington

Notably, open-source models were also discussed during this meeting, with representatives from the renowned chip manufacturer NVIDIA attending. Previously, Huang Renxun, CEO of NVIDIA and an advocate for open source, visited Washington last week to meet with U.S. Commerce Secretary Howard Lutnick and the President.

Controversies surrounding Chinese open-source AI models continue to intensify—factions led by OpenAI and Anthropic are trying to push for restrictions on the use of Chinese open-source models, while giants like NVIDIA, Microsoft, and Meta have jointly opposed this. By incorporating open-source issues into the AI evaluation framework discussion, the White House indicates that this debate is rising from the industry level to the policy decision-making level. Both the White House and the EU have refused to comment on the secrecy of the framework, and the UK government has also not responded. When even participating companies are kept in the dark about the regulatory rules that the world's most powerful AI models will face, this "completed but unseen" framework may already reveal much.