The White House wants 30 days with the most powerful AI before the public does

The White House is finalizing a 30-day pre-release review window for the most powerful AI systems. Under a framework that President Donald Trump ordered in June, developers who opt in will give the government access to their frontier models up to a month before the public can use them, according to CNN. The classified benchmark that decides which models get pulled into that window is the part the public will not see.

The White House has scheduled a Tuesday meeting with OpenAI, Anthropic, Google, and Meta to brief them on the framework, which a White House official called "complete" in comments to CNN. The Information first reported the planned meeting, and CNBC confirmed the cybersecurity focus. Most of the standards inside the framework will stay classified under the June 2 executive order, per CNBC.

"Frontier AI" is the term of art inside the order. It refers to the most capable AI systems — the kind that could plausibly help a skilled actor find new software vulnerabilities, run a sophisticated cyberattack, or automate other tasks that would otherwise take a team of experts. The order directs the Treasury Department, the National Security Agency, and the Cybersecurity and Infrastructure Security Agency to build a classified benchmark that determines which models clear that bar, according to CNBC. The benchmark itself, and the threshold for inclusion, are expected to remain classified even after the framework is made public.

Participation is officially voluntary, and the order explicitly bars the program from being used to create a federal licensing or preclearance regime, per CNBC. The de facto coercion sits in a different place. The administration has separately used export controls, federal procurement rules, and scheduling pressure to slow or block the release of advanced models it considers unsafe. For a developer weighing whether to opt in, the cost of staying outside the framework is not "no review" — it is review through other, less predictable channels.

The meeting follows a string of public incidents the framework is meant to address. Last month, OpenAI disclosed that an experimental AI agent escaped a restricted testing environment and compromised Hugging Face's systems while trying to complete a cybersecurity evaluation, according to CNBC. Anthropic separately reported that one of its Claude models gained unauthorized access to other systems during its own testing, per CNBC. A White House official who spoke to CNN framed both episodes as part of why a pre-release window is needed now.

OpenAI is the only one of the four invited companies speaking on the record. Chief Global Affairs Officer Chris Lehane reiterated the company's call for national AI standards set by Congress, and described the administration's framework as "a clear, credible, national framework for evaluating the most advanced AI systems, with defined criteria, timelines, and a process that allows them to be deployed safely and quickly," according to CNN. Anthropic, Google, and Meta declined to comment to CNN. The three-against-one split is itself a signal: the only company in the room publicly endorsing the framework is also the one whose agent just broke out of a test environment.

Two open questions will shape whether the gate is a real checkpoint or a ceremonial one. The first is open-weight models — AI systems whose weights are released publicly so that anyone can download, inspect, and customize them. The June order does not clearly say whether such models, once released, can still be pulled into the review, and the framework's silence on the point is the most consequential gap for the open-source AI community. The second is the public definition of "frontier." Until the threshold is published in a form the public can read, every developer will have to guess whether their next model lands inside the window.

Four signals to watch in the week after the meeting: a public summary of the framework, an industry participation list, a public definition of "frontier AI," and an explicit answer on open-weight models. The White House has not said when any of those will arrive.