White House Completes New AI Framework. Its Rules Are Still Not Public

The White House has completed a long-awaited voluntary framework for evaluating the nation's most advanced artificial intelligence models, meeting a deadline established by President Donald Trump's June executive order.

However, the administration is keeping nearly every aspect of the framework under wraps. "The voluntary framework outlined in the June 2nd executive order was complete by the deadline," a White House official told Axios Monday, adding that discussions with industry about implementation are already underway.

The official also said the administration is working with more AI companies than just OpenAI, Anthropic and Google, which previously reviewed a draft version before its completion. The framework stems from President Trump's June 2 executive order, which directed federal agencies to create a voluntary process allowing AI developers to determine whether models under development qualify as "covered frontier models."

Those models could then be shared with the government for up to 30 days before public release under strict confidentiality, cybersecurity, and intellectual property protections. According to the executive order, developers participating in the program would also work with the federal government to identify "trusted partners" that could receive early access to advanced AI systems in an effort to strengthen cybersecurity and protect critical infrastructure.

The order specifically states that participation is voluntary and does not establish a licensing or preapproval requirement for AI development. Despite those broad outlines, the administration has declined to release the actual framework or explain how developers will determine whether their models fall within its scope.

One of the most closely watched elements is the government's classified benchmarking process, which will evaluate whether advanced AI models possess significant offensive cyber capabilities.

Under the executive order, both the testing methodology and the threshold that determines which systems qualify as covered frontier models are classified and will only be shared with developers and researchers when appropriate.

The lack of public guidance creates uncertainty for AI developers racing to build increasingly capable models. Companies want to know as early as possible whether systems currently under development will trigger the government's voluntary review process, allowing them to plan product launches and compliance strategies before release.

The framework also arrives as Washington becomes increasingly focused on the cybersecurity implications of frontier AI. Recent demonstrations showing advanced models performing sophisticated hacking tasks have intensified concerns inside government about protecting critical infrastructure while avoiding regulations that could slow U.S. innovation.

Industry engagement is expected to accelerate this week. According to a source familiar with the plans, White House officials are scheduled to hold a staff-level meeting Tuesday with AI companies to review the framework and discuss how it will operate in practice. Representatives from OpenAI, Anthropic, Google, Meta and other leading developers have been invited to participate.