White House finalises its frontier-model framework — then declines to publish it
The administration reviewed its finalised voluntary framework for evaluating frontier models with Meta, Nvidia, Microsoft, OpenAI, Anthropic and a range of smaller companies on 4 August — and then said the details will not be released publicly. Under Executive Order 14409, participating developers can give federal agencies secure access to covered models for up to 30 days before release.
A voluntary framework works by reputation. Companies participate because participation is visible and non-participation is visible, and outside observers can tell which is which. Withhold the contents and that mechanism stops functioning: no external party can evaluate whether the bar is demanding or cosmetic, and no participant can be held to a standard nobody outside the room can read.
The definitional gap makes it sharper. The executive order does not specify the technical criteria that make a model a covered frontier model, and the criteria that would have filled that gap are precisely what is now being withheld. The scope of the regime and the substance of the regime are both non-public, which leaves the question of who is covered answerable only by the people doing the covering.
There is a legitimate case for some opacity — capability thresholds are genuinely dual-use information, and publishing an exact bar tells a bad actor exactly how to sit beneath it. That argument covers specific numbers. It does not cover the process, the participant list, or the compliance criteria, and running the whole thing dark is a choice rather than a necessity.
Fortune — White House won't publicly release AI model evaluation framework it reviewed with OpenAI, Anthropic, Microsoft and others → · CNBC — White House to host AI companies Tuesday to review new model-testing framework → · Tech Policy Press — Five questions the US government should answer about its secretive frontier AI framework →