White House unveils voluntary framework for pre-release review of closed models

The United States administration has placed on the table a voluntary framework that permits developers of closed-source AI models to submit them for a government review thirty days before a public launch, with the assessment conducted in an isolated environment that enforces strict access controls and logs every user. A blanket exemption is granted to models with open weights; the rules apply only to closed models and are not intended to limit the free download of weights after release.

OpenAI, Anthropic, Google, Meta, Nvidia and other large firms were invited to the presentation, and the review is expected to be carried out in cooperation with the NSA and additional agencies concerning national-security risks. However, many details remain absent: the definition of what constitutes a “frontier” model (a cutting-edge AI system) has not been formulated, the criteria for national-security threats have not been published, the document itself will not be disclosed to the public, the list of trusted partners stays confidential, and small developers are left out of the process without knowledge of the rules of the game.

The major companies have asked to receive near-final versions rather than early drafts, effectively creating a direct channel between the large laboratories and the government while the rest of the market operates in opacity. The outcome is a built-in asymmetry: those already possessing resources and access obtain regulatory certainty, whereas those without remain in complete uncertainty.