The White House has announced that its AI framework is complete, but it has not disclosed its contents.

      TL;DR: The White House has finished its voluntary AI framework on schedule but will not reveal its details. The benchmarks and model thresholds are classified. OpenAI, Google, and Anthropic reviewed a draft. A staff-level meeting is set for Tuesday.

      On Monday, the White House announced that it met its deadline for completing a voluntary framework to assess advanced AI models. However, it will not disclose the framework’s contents, who has access to it, or when companies can implement it. “The voluntary framework outlined in the June 2nd executive order was completed by the deadline,” a White House official stated. “Discussions with the industry regarding the next steps are ongoing.”

      The framework provides the government with a system to determine if AI models in development fall under the June executive order, which established a 30-day review period before the release of frontier models. The benchmarks for evaluating cyber capabilities are classified, and the criteria for which models are included are likewise classified, shared only with developers “as appropriate.” While the framework itself is not classified, the White House is withholding it from the public. “Just because things are unclassified doesn’t mean we will broadcast them to everyone,” said the official.

      OpenAI, Anthropic, and Google provided input on a draft version of the framework. The administration claims to be consulting with “many more” industry partners beyond these three companies. A staff-level meeting to discuss the completed framework is scheduled for Tuesday. Although Trump’s June executive order characterized the review process as voluntary, the combination of classified benchmarks, undisclosed thresholds, and a 30-day government review period effectively creates a barrier that companies cannot publicly assess or contest.

      Policymakers and AI safety advocates were expecting to see specifics but have not received any. The White House launched Gold Eagle this month to coordinate AI-driven cyber defense, and the model evaluation framework serves as its complementary component: Gold Eagle identifies vulnerabilities, while the framework determines which models are substantial enough to warrant government review prior to their release. The critical issue is whether a framework that is inaccessible to those outside the government, based on benchmarks that are not available for public examination, meets the transparency standards promised by the executive order.