Major players in the artificial intelligence industry met with the U.S. government this week in an effort to resolve mounting confusion over the regulation of frontier AI models. The meeting, convened at the White House on Tuesday, brought together OpenAI, Anthropic, Google, Meta, Nvidia, and other leading AI companies. The outcome was a new voluntary framework permitting government review of frontier models before public release.

However, the companies that participated in the White House gathering agreed to an arrangement the general public will not be permitted to see.

The administration does not plan to publish the framework it has spent the last two months developing, following President Trump's June order directing officials to create it. Key details—including which models are covered, the relevant thresholds, and the list of "trusted partners" granted early access to the world's most powerful models—remain unknown to the public and much of the industry. In the absence of comprehensive AI legislation from Congress, executive actions and voluntary industry agreements remain the primary instruments of U.S. AI oversight, a gap that contrasts with the European Union's phased rollout of its legally binding AI Act.

The June order already stipulated that the benchmarking process used to designate a "covered frontier model" may be classified, and that the determination would rest with the director of the NSA.

A source familiar with the situation said only a handful of companies were present in the briefing room when the framework was discussed. This underscores a broader concern that U.S. AI regulation is increasingly being shaped through conversations among a small number of dominant firms. Smaller and open-source labs worry these safety frameworks could influence broader market structure and entrench the use of closed-source frontier models that carry a "government-approved" designation.

What's Actually in the Framework?

Reports indicate that models covered by the framework are defined as closed-source, demonstrating state-of-the-art capabilities and presenting national security risks. However, a person familiar with the briefing said neither "state-of-the-art" nor "national-security risk" has been clearly defined.

Developers can voluntarily submit such a model to the government for up to 30 days before release. Companies were reportedly instructed to submit models as close to launch-ready as possible rather than at early checkpoints—roughly where things stood in July. The review will be conducted by an assortment of administration officials rather than a single agency.

Based on accounts from sources familiar with the process, open-weight models appear to have been excluded from the framework. Some observers note this exclusion could benefit companies attempting to catch up with industry leaders, as they would face fewer regulatory obstacles. At the same time, the exclusion would appear to leave outside of oversight precisely the AI models Washington is most concerned about: open-weight releases from Alibaba, DeepSeek, and Moonshot AI that continue to approach the American frontier.

Even companies working with open weights remain unsure whether their models could be submitted, or what that process would entail. Some argue that open-source models not being explicitly included could ultimately harm the companies building them, as it may push customers toward "government-approved" closed-source alternatives.

Another notable detail: during the reported 30-day review window, submitted models are to be held in high-security environments where access is logged in detail and, per Axios, "employees would be limited from accessing models." This appears to suggest that a company's own staff would be restricted—or at least limited—from using its own frontier model while Washington evaluates it. Internal deployment, in which companies use an unreleased model themselves, has been cited as a blind spot in many previous governance proposals.

The issue is especially topical given that the recent hacks carried out by OpenAI's escaped autonomous agents were reportedly facilitated in part by a secret unreleased model.

Critics Say the Rules Remain Unclear

Criticism of the framework and the government's approach has come from multiple directions.

"This is not a handshake deal with tech companies. It's the rulebook for ensuring they don't endanger the public. If only tech companies know what's in the rulebook, it doesn't work," Americans for Responsible Innovation, a Washington-based AI policy nonprofit, said in a post on X. The organization has been advocating for more transparent, enforceable federal rules around advanced AI systems.

Adam Thierer, a resident senior fellow in technology and innovation at the R Street Institute, said the administration "appears destined to give us something far more arbitrary and burdensome" than its predecessor from the Biden administration "with this behind-closed-doors de facto licensing regime they are concocting." The Biden administration's 2023 executive order on AI had required developers of the most powerful foundation models to share safety test results with the government, though it too stopped short of establishing a formal regulatory regime.

Rep. Lori Trahan, co-sponsor of the new bipartisan FRONTIER Act, which would place frontier AI oversight within a civilian-led federal framework, argued on X that AI governance "belongs in a civilian agency, where it can be seen and questioned, not buried inside the national security apparatus."

Until the rulebook is brought into the open, critics warn, Washington may be quietly deciding who gets a head start in the AI race.

By Beatrice Nolan for Fortune.