White House AI evaluation framework will remain confidential

The White House has decided not to publicly release its newly completed framework for evaluating advanced AI models, according to three sources familiar with the discussions, as first reported by Axios.

The voluntary framework, which establishes procedures for AI developers to work with the government in assessing whether their models meet certain thresholds, will only be shared with companies participating in the process. Staff-level meetings with industry representatives took place Tuesday to review the framework's contents, while companies not invited to those sessions remain unaware of the details.

Why it matters

The decision to keep the framework confidential creates an information asymmetry in AI governance. Researchers, policymakers, and U.S. allies working on AI safety will lack visibility into how the administration plans to implement one of its central AI policies, even as the framework carries global implications for AI security standards.

What the framework covers

The framework stems from a June executive order that outlined a voluntary process for AI developers. Under the system, companies can work with the government to determine whether their models fall within the framework's scope before release.

The framework is expected to establish rules for government access to AI models up to 30 days before their public release. These rules will address confidentiality protections, cybersecurity measures, insider-risk protocols, intellectual property safeguards, and nondisclosure requirements.

The executive order explicitly states that the benchmarking process used to assess advanced cyber capabilities of AI models will be classified. The order contains no requirement to publicly release the voluntary framework itself.

Open questions remain

Several aspects of the framework's implementation remain unclear. Sources indicated that open-source models were discussed during Tuesday's meetings, though they did not provide specifics. Nvidia staff participated in the sessions, and CEO Jensen Huang, a vocal open-source advocate, met with Commerce Secretary Howard Lutnick and President Trump in Washington last week.

It's also uncertain which "trusted partners" will receive early access to advanced models under the framework. Whether any foreign governments would qualify for this designation remains unknown. The European Union declined to comment on the matter, and the U.K. did not respond to multiple requests for comment.

Timing and context

The framework arrives as the AI industry confronts high-profile cyber-attacks and rapid advances by Chinese AI developers, reigniting debates over how to handle open-source models and their security implications.

The White House declined to comment on the framework's confidential status.

Details of the confidential framework were first reported by Axios, with reporting by Maria Curi.