Washington Finalises Frontier Model Review Framework, Keeps Rules Private

The US administration has completed a framework intended to govern the government's early access to the most powerful AI models. But it has not published the text, identified the companies that have seen it or said when it will be used. Meeting an official deadline has therefore raised a larger transparency question: how can Washington's allies, model developers and the public assess a rule that affects technology releases when the rule itself remains behind closed doors?

A White House official told Axios that the voluntary framework set out in the June 2 executive order had been completed by the deadline, and that discussions with industry about next steps were continuing. The official added that the administration was engaging with more partners than Anthropic, OpenAI and Google. The three labs had provided feedback on a draft of the framework before the deadline expired.

Thirty days could reorder the moment of release.The framework was designed to give developers a channel for communicating with the government and determining whether a model under development falls within the category of covered advanced models. If it does, the framework is expected to set out confidentiality, cybersecurity, insider-risk, intellectual-property and non-disclosure terms when the government receives access to the model for up to 30 days before it becomes available to other trusted partners. It is also expected to address how the partners receiving that early access are selected.

The problem is that the threshold for entering this process is not public. The executive order states that the benchmarking process used to assess models' advanced cyber capabilities will be classified. The threshold that brings a model within scope will also be classified, and shared with developers and researchers only as appropriate. The order did not describe the voluntary framework itself as classified, so policymakers and observers had been waiting for details. The US official's response was that information not being classified does not mean the government will broadcast it to everyone.

A voluntary process does not eliminate the state's effect on the market.The report provides no evidence that the framework is a mandatory licence, or that Washington now legally approves every release. But it reveals a structured channel that gives the government early visibility into capabilities not yet on the market, and a role in identifying the trusted partners that may receive access before others. For companies, early clarity matters. They need to know whether a model under development will enter the process and what safeguards will protect their secrets and intellectual property.

Those who read the rule late start the race late.Viewed through the lens of sovereign access to models, this is not only about US security. The most powerful systems are built by American companies and reach institutions and governments abroad through APIs, cloud services and partnerships. If an early-access window becomes an established part of a release, the definition of a trusted partner and the way evaluation findings are shared could affect who tests the capabilities first and who waits for the public version.

This is where the stakes lie for the Middle East and North Africa. Governments, cloud operators and regional labs that depend on US models need to know whether availability schedules, cyber capabilities or terms of use could change because of a review whose rules they cannot see. That does not mean assuming a delay or exclusion the source did not report. It means that dependence on a single supplier adds a political and operational variable that does not appear on a price list. The risk can be reduced by evaluating several models, maintaining a migration plan, and building local tests for Arabic and security instead of waiting for a complete external judgement.

The White House is expected to hold a staff-level meeting with companies on Tuesday to review the framework. But even that meeting will not answer the public questions: what characteristics bring a model within scope? Who oversees the protection of secrets during government access? How is a trusted partner selected? Will any findings or principles be published so allies and markets can understand the process's effect?

The honest conclusion is that Washington has completed the procedural structure requested by the executive order, according to its official, but has not yet provided enough information to assess its effectiveness or the fairness of its application. Secrecy may be understandable when testing sensitive offensive capabilities. It becomes more problematic when it conceals the rules for dealing with industry. The framework's value now depends not on its completion on paper, but on the clarity that emerges after the company meeting and its ability to protect security without turning early access into a closed circle.