Representatives from roughly a dozen AI companies, including OpenAI, Anthropic, Google and Meta, met with White House officials on the morning of August 4 to close out nearly two months of negotiation over a voluntary framework governing how the federal government reviews frontier AI models before release. Officials confirmed after the 30-minute meeting that the framework is now final, but declined to disclose its contents, who has reviewed it, or when companies will begin using it.

What the framework does

The process traces back to a June 2 executive order, "Promoting Advanced Artificial Intelligence Innovation and Security," which directed the Treasury, Homeland Security and Defense departments — in consultation with NIST and the White House science office — to design an opt-in system letting developers give the government early access to "covered frontier models" for up to 30 days before wider release, down from a 90-day window in earlier drafts. The order also directed Treasury, the NSA and CISA to build a classified benchmark for judging which models' cyber capabilities are significant enough to trigger review. The executive order explicitly bars the framework from becoming a mandatory licensing or preclearance regime, and reporting indicates the reviewing body will draw on NIST's Center for AI Standards and Innovation (CAISI) even though CAISI is not named in the order itself.

Confidentiality and criticism

The White House's decision to keep the finished framework private drew immediate pushback: companies that weren't part of the negotiations have no visibility into what they'd be agreeing to, and there is no mandatory breach-reporting requirement built into the voluntary system. The timing has also raised eyebrows — the meeting came within days of OpenAI's and Anthropic's own disclosures that their models breached real systems during internal cyber testing, and Anthropic, which is currently suing the administration over an unrelated Pentagon contracting dispute, was nonetheless invited to help shape the government's frontier-model safety process.

Why it matters

This is the US government's first concrete step toward a working relationship with frontier labs on pre-release safety review, following months of voluntary commitments with little enforcement teeth. Whether a framework negotiated and kept confidential by the same handful of companies it applies to meaningfully changes how frontier models are tested before shipping — as opposed to formalizing an process labs were already doing informally — will depend on details the administration isn't yet willing to share.