White House finalizing 30-day pre-release reviews of frontier AI models

AI regulation is getting real: the US government wants to see every new frontier model before you do.

The framework taking shape

The White House is finalizing a voluntary framework with OpenAI, Anthropic and Google that would give federal agencies up to 30 days to review a new frontier model’s national security implications before public release. The reviews would run on classified benchmarks, and an announcement is expected before August 1, 2026.

The legal basis is Executive Order 14409, signed June 2, 2026. Its structure: a 30-day pre-release window, classified evaluation criteria, and voluntary participation. There is no licensing and no mandatory approval, but the labs that matter are signing up anyway.

The Meta-shaped hole

The loudest part of the story is who is missing. Meta has not joined the framework, and reporting says the White House is now applying direct pressure on it to sign.

That standoff will define how “voluntary” this framework really is. If the last major holdout can be pressured into joining, the distinction between a voluntary program and a de facto requirement starts to look academic. Critics point out exactly this tension: a 30-day government preview of every major model release is a big step for a framework that still calls itself voluntary.

What this changes

The quotable version: under EO 14409, OpenAI, Anthropic and Google would give US agencies up to 30 days of pre-release national security review of new frontier AI models.

For model release calendars, that is a real new constraint. Launch dates would carry a built-in government review window, and classified benchmarks mean the public will not know what the models were tested against. Whether early government access to frontier models is prudent oversight or a quiet precedent depends on where you sit; either way, the announcement is expected to land within days.