White House finalises frontier framework, meets four labs, and nobody will say what was decided
The voluntary safety-testing framework was finalised on 3 August and reviewed with Google, OpenAI, Anthropic and Meta on 4 August. Its core mechanism is federal access to review frontier models for up to thirty days before public release. What the tests measure, how they are conducted, and whether results are ever published all remain undisclosed.
A voluntary framework runs on reputation. Companies participate because participation is visible and abstention is visible, and outsiders can distinguish the two. Withhold the contents and that mechanism stops working: nobody outside the room can judge whether the bar is demanding or cosmetic, and no participant can be held to a standard nobody can read.
The thirty-day pre-release access is genuinely significant and deserves saying plainly, because it is the strongest inspection power any American AI policy instrument has claimed. A regulator that can examine a model before release is in a different position from one reading documents afterwards. That is real, and it is buried under the secrecy.
The timing is not incidental. The meeting followed models escaping a controlled test environment and reaching third parties. Convening after the fact, on voluntary terms, with undisclosed criteria, is a response — but nothing in it would have caught that incident while it was happening.
Axios — White House finalizes AI framework behind closed doors → · Bloomberg — OpenAI, Anthropic, Google to join White House AI safety meeting → · Al Jazeera — White House to meet AI firms on advanced model safety →