The White House said Monday that the United States had completed a voluntary framework for government evaluation of the most advanced artificial intelligence models, meeting a 60-day deadline set by President Donald Trump's June 2 executive order. The administration has not publicly released the framework or explained which models will qualify, while discussions with AI companies about implementation are continuing, according to Reuters and Axios.
The framework creates a channel through which developers can ask federal agencies whether a model under development meets the classified threshold for a covered frontier model. Under the executive order, participating companies may give the government access for up to 30 days before releasing a qualifying system to other trusted partners. The order explicitly rules out mandatory licensing, preclearance or permits for new AI models.
The White House order assigns the National Security Agency, the Cybersecurity and Infrastructure Security Agency, the National Institute of Standards and Technology and other officials roles in the process. It directs them to maintain classified benchmarks for advanced cyber capabilities and to protect confidentiality, cybersecurity, insider-risk controls and intellectual property when agencies examine unreleased technology.
A White House official said the framework was finished by the deadline and that talks with industry on the next steps were under way. Axios reported that Anthropic, OpenAI and Google had provided feedback on a draft, while officials were engaging with additional industry partners. A staff-level meeting with technology companies was expected Tuesday, but the administration had not disclosed the framework's text, participation timetable or full roster of companies.
The review system is intended to identify powerful cyber capabilities before public deployment while preserving rapid commercial development. It follows heightened concern about AI systems finding software vulnerabilities or acting in unexpected ways. The wider June order also called for an AI cybersecurity clearinghouse to coordinate vulnerability discovery and patches, and for defensive tools to reach federal agencies, critical infrastructure, rural hospitals, community banks and local utilities.
The immediate test will be whether developers join the voluntary process and whether the government supplies enough detail for companies to know when a model is covered. Policymakers, researchers and US allies are also waiting to see how early access will work and how sensitive model information will be secured. Federal officials have said the capability benchmarks and coverage threshold remain classified, leaving implementation and industry participation as the next measurable steps.
Comments