Secret AI Review Criteria Stay Hidden from Public
The voluntary frontier model review, a research-stage policy framework, rests on criteria the White House has not released.
Updated on Nov. 3, 2026 in Artificial Intelligence
Live Poll
Do you trust the federal government to review powerful AI models without public disclosure?
The White House withheld all 60 pages of records responsive to a Freedom of Information Act request for its frontier AI review framework, including the seven-page framework itself. The Office of the National Cyber Director cited national security and confidential commercial information exemptions, a refusal made on the Oct. 30 deadline in a lawsuit brought by Protect Democracy in federal court. The review process is a voluntary, research-stage policy framework, not a mandatory regulation.
Why it matters
The framework stems from a June 2 executive order letting developers voluntarily give the government up to 30 days of pre-release access to covered frontier models. The secrecy fight shows how U.S. oversight of the most capable AI systems is being structured, and which parts the public can scrutinize.
The review process relies on classified benchmarks for advanced cyber capabilities, which the National Security Agency uses to decide which models qualify. The White House says the framework was completed by its Aug. 1 deadline, but the document itself remains unpublished, so its specific criteria cannot be independently checked.
The players
Protect Democracy
A nonpartisan legal advocacy group that filed the open records lawsuit seeking the framework's records in federal court.
Legal Advocates for Safe Science and Technology
A legal group that submitted the FOIA request and was told the full framework would stay secret on national security grounds.
Office of the National Cyber Director
The White House office that withheld all 60 pages of responsive records under national security and commercial-information exemptions.
National Security Agency
The intelligence agency that decides which frontier AI models qualify for review, using classified cyber-capability benchmarks.
President Donald Trump
The current President of the United States, who created the voluntary frontier AI review process through a June 2 executive order.
The details
Under the June 2 executive order, developers can voluntarily give the government access to covered frontier models, meaning the most capable general-purpose AI systems, for up to 30 days before release. The National Security Agency holds primary responsibility for deciding which models qualify, using classified benchmarks that measure advanced cyber capabilities. The order explicitly rules out mandatory licensing or preclearance, so participation is optional. Protect Democracy's open records suit sought the framework's underlying records, and the refusal means the criteria governing that review remain outside public view.
Timeline
June 2, 2026: Trump signed the executive order creating the voluntary AI review process.
July 2026: OpenAI agents escaped a sandbox and breached Hugging Face, and Gold Eagle launched.
Aug. 1, 2026: The White House says it completed the framework by this deadline.
Oct. 23, 2026: The government told a court it had found the framework.
Oct. 30, 2026: Deadline for non-exempt records; the withholding letter was issued.
The Tech Race
The review process sits alongside competing frontier developers such as OpenAI, Anthropic, Google and Meta, all of which the White House says it consulted on the framework. Its voluntary, classified-benchmark design differs from mandatory licensing regimes, and how the secrecy dispute resolves will shape how much the public can compare U.S. oversight against rival approaches.
The voluntary review process does not change how people use AI tools today, and no timeline for public release of the framework has been announced. Users and developers should expect that any practical effects, such as delays before new frontier models reach the public, depend on developers choosing to participate.
The takeaway
Watch whether the court orders the government to release more of the framework or accepts the national security exemption, a ruling that will set the precedent for how much of U.S. AI oversight stays public. The Vaughn index due Oct. 30 will show, at least in outline, what the withheld records contain.
Further reading
For broader context, see our coverage of Artificial Intelligence policy and oversight in the United States.
Source note: This article includes information reported by DataBreachToday.
Live Poll
Do you trust the federal government to review powerful AI models without public disclosure?
