The White House says it completed a voluntary framework for government evaluation of advanced AI models, but it has not published the text, coverage threshold or implementation schedule. OpenAI, Anthropic and Google provided feedback on a draft before a staff-level industry meeting. The White House said the framework was completed by the deadline in a June executive order. The administration did not release the framework text.
OpenAI, Anthropic and Google gave feedback on a draft. The framework is intended to govern model access for up to 30 days before release. The cited reporting attributes statements to the officials or organizations making them and keeps those claims separate from events independently observed or documented.
Voluntary participation differs from a binding regulation. Model developers need advance notice to know whether a system is covered. Classified technical thresholds do not automatically require the entire governance process to remain secret. That background explains the stakes without resolving the decisions or outcomes still pending.
Confidentiality, cybersecurity, insider risk and intellectual-property rules are among the expected subjects. The model-coverage threshold and cyber-capability benchmarks are classified. Dates, counts and legal status remain tied to the source record because later updates may revise preliminary information.
The evidentiary limit is specific: The number of participating companies, start date and decision authority were not publicly confirmed. This report therefore does not infer motive, causation, final totals, legal outcome or implementation beyond the available evidence.
The next concrete developments are any public summary or company participation notice and evidence of the first model review under the framework. Those records will show whether the reported development changes policy, operations or public risk.
Taken together, the verified record is narrower than the broadest claims surrounding the story. The White House said the framework was completed by the deadline in a June executive order. The administration did not release the framework text. OpenAI, Anthropic and Google gave feedback on a draft. The framework is intended to govern model access for up to 30 days before release. Confidentiality, cybersecurity, insider risk and intellectual-property rules are among the expected subjects. The model-coverage threshold and cyber-capability benchmarks are classified. The context is equally important: Voluntary participation differs from a binding regulation. Model developers need advance notice to know whether a system is covered. Classified technical thresholds do not automatically require the entire governance process to remain secret. This synthesis preserves what is known while keeping the stated limits visible.