White House finalises a 30-day pre-release review for frontier models

The White House is finalising a framework requiring a 30-day pre-release review of frontier models, to take effect before August 1.

OpenAI, Anthropic and Google are participants. Meta is excluded. The benchmarks used in the review are classified, meaning neither the public nor the excluded labs can see what the models are tested against or what constitutes passing.

The review would sit between a model being completed and being released, adding a fixed delay to launch timelines for participating labs.

Separately, AMD may invest up to $5 billion in Anthropic, released against deployment milestones. AMD already holds partnerships with OpenAI and Meta, giving it commercial relationships across most major labs as it competes with Nvidia in AI infrastructure.

Two details limit what the framework can achieve. Classified benchmarks cannot be independently reviewed or contested, and Meta's exclusion leaves the most widely self-hosted open-weight family outside the process.