White House finishes AI framework, keeps its contents secret

The White House says the voluntary AI framework spelled out in Trump's June 2 executive order is done. It also says it will not tell the public what is in it. That combination is the story The Next Web landed this week.

The bones that have been made public are these. The framework gives the government a 30-day preview window on frontier models. The benchmarks used to assess cyber capability are classified. The thresholds that decide which models fall under review are classified. The framework document itself is unclassified but is not being released, on the logic, offered by a White House official, that 'just because things are unclassified that doesn't mean we are going to broadcast them to everyone.' OpenAI, Anthropic and Google were shown draft versions, and a staff-level meeting with companies to walk through the finished text was reportedly set for Tuesday.

If you run policy or product at a frontier lab, the interesting part is not the secrecy for its own sake. It is that a rule you cannot read is a rule you cannot plan around. If your next training run may or may not cross a classified threshold that only the NSA can actually see, the safe move is either to be one of the three companies already in the drafting room, or to slow down and ask. Either outcome concentrates advantage. The companion Gold Eagle initiative for AI-powered cyber defense, launched this month, sits next to the framework in the same architecture: the government wants a channel into frontier releases, on its own terms.

The honest caveat is that the reporting here is thin on specifics precisely because the specifics are being withheld. What the piece does not tell you is what actually starts the 30-day preview clock, whether open-weight releases or non-US labs sit inside the scope, or whether a company told its model has crossed a classified line has any route to contest that call. Those are the parts a general counsel would want in writing before signing off on anything.

What is worth watching is whether the three labs in the room end up quietly speaking for the industry, or whether smaller US developers and the open-source community push for the framework text itself to be released. A voluntary regime only stays voluntary as long as the people it applies to consent to play.