THE ESSENTIALS IN 30 SECONDS

  • The Office of the National Cyber Director declined on October 30 to produce key records on the voluntary AI model testing framework, The Information reported.
  • Reports on the letter say it cited national security and commercial exemptions; the framework is described as a seven-page document.
  • Protect Democracy Project had sued in September and obtained an agreement that records, plus an index of any exempt material, would be produced by October 30.

The White House has refused to release the main documents behind its voluntary system for testing advanced AI models before launch. According to The Information, the Office of the National Cyber Director (ONCD) sent a letter on Friday, October 30, to the nonprofit Protect Democracy Project declining to produce the key records, even though the government had agreed in September to hand them over by that date.

What was requested and what was withheld

Protect Democracy filed a Freedom of Information Act request in August and sued on September 1 in federal court in Washington. The defendants are ONCD, the Department of Commerce, the Treasury Department and the White House Office of Science and Technology Policy. In a joint status report filed on September 15, the group narrowed its request to four items: substantively unredacted text of the framework, records given to AI companies on the terms of taking part in a clearinghouse called GOLD EAGLE, documents identifying the "trusted partners" under the review process, and the legal authority used to set it all up. The government agreed to produce records by October 30 and to provide a Vaughn index, a log justifying any withheld passages.

Reports summarizing the letter say ONCD invoked national security and commercial-confidentiality exemptions. Posts by The Information's reporter Leo Schwartz described the framework as a seven-page Word document and said ONCD and Commerce appear to be approving trusted partners for pre-release access to models. The letter itself has not been published, and it is not yet clear which records were withheld in full or in part. The next joint status report is due on November 13, when the parties may set a briefing schedule.

The framework and the labs

The process stems from an executive order signed on June 2 that gave officials 60 days to build a voluntary review for the most capable models. Under it, developers may give the government access to covered models up to 30 days before wider release, and the order rules out mandatory licensing. The White House said the framework was finished by its August 1 deadline but did not publish it.

Participants have not been officially disclosed. A White House official said in August that OpenAI, Anthropic and Google gave feedback on a draft, and CBS News reported in September that OpenAI chief executive Sam Altman said the company had submitted a model for review. Protect Democracy argues that the government has described the framework only in general terms, leaving its standards, participants and legal basis unknown to the public.