Five outlets, Bloomberg, CNBC, CNN, Axios and The Next Web, reported on August 3 that the White House has finished the AI framework mandated by President Trump’s June 2 executive order, “Promoting Advanced Artificial Intelligence Innovation and Security.” The administration would not say what the framework contains, who has read it, or when companies are expected to start applying it.
What the June 2 Executive Order Actually Requires
The order directed federal agencies to build a framework, due August 1, covering how developers of frontier AI models interact with the government before a model ships. Participation is voluntary, and the order explicitly rules out mandatory licensing or a pre-clearance requirement.
Deciding which models are covered isn’t public either. The threshold runs through a classified process. The same order directed federal agencies to develop and maintain a classified benchmarking process for evaluating a model’s advanced cyber capabilities.
What Is Known and What Is Classified About the White House AI Framework?
The June 2 executive order directed agencies to produce a framework by August 1 covering how frontier AI developers deal with the government before a model launches. Participation is voluntary: developers who opt in give the government early access to a model for up to 30 days ahead of a broader release, and the order rules out mandatory licensing or pre-clearance. Reporting on August 3 established that the framework is finished. What stayed closed is just as specific. The order placed the benchmarking methodology used to evaluate a model’s cyber capabilities under classification, along with the threshold that decides which models count as “covered frontier models,” a figure set through a classified process and shared with developers only “as appropriate.” Also undisclosed: who outside government has reviewed the finished text, and when companies are expected to start applying it.
The White House’s Own Explanation for the Secrecy
Asked why the framework is being withheld from public view, a White House official gave The Next Web a direct answer.
“Just because things are unclassified that doesn’t mean we are going to broadcast them to everyone.”
— a White House official, via The Next Web
That line covers the framework document itself, which the White House has not released. The benchmarking methodology and the model threshold sit in a different category: the June 2 order placed those under classification directly. The threshold reaches developers only “as appropriate,” according to reporting on the framework.
A Staff-Level Meeting, With Industry Already in the Draft
Administration staff meet with AI companies on August 4, a session that includes OpenAI, Anthropic and Google among others. Those three companies had a hand in the text: they reviewed a draft and sent edits in late July. Press reporting expects the finished framework to spell out obligations around confidentiality, cybersecurity, insider risk and intellectual property protection during the government’s review window, though the administration has not confirmed that content directly.
What This Means for Anyone Planning Around Model Releases
None of this changes what a model can do. It changes the shape of the calendar for anyone who opts in. For a developer that opts in, a public release carries an early-access window worth up to 30 days somewhere ahead of it, triggered by a threshold that reaches developers only as appropriate. Teams planning a launch or an integration around an assumed release date are planning against a variable they cannot see.
Control over frontier models keeps taking forms that are easy to miss. Moonshot AI’s revenue-tiered license on the Kimi K3 weights attaches obligations to a dollar threshold, one a licensee can read in the license text and plan against. The White House threshold works the other way: it decides who is covered without being published. Publishers weighing whether to block Google’s crawlers hit a related wall, where every control that reaches AI Overviews carries its own cost, and none leaves a publisher indexed, snippeted and excluded from AI Overviews at once.
What holds up under the reporting is narrow but solid: the framework exists, it traces back to the June 2 order, and participation in it is voluntary. How a model gets flagged for review, who outside government has read the finished text, and when companies are expected to start using it remain outside what the White House will confirm.
Sources: The Next Web’s report on the finished framework; Skadden’s analysis of the June executive order.