The Secret AI Model Review Rules: A Call for Transparency
The rules governing the pre-deployment review of frontier AI models are shrouded in secrecy. Only a select few large companies have had access to these guidelines.
More than two dozen organizations, spanning the political spectrum, have united to demand the White House publish this voluntary framework. This move comes after smaller and open-source developers were largely left out of the initial preview.
September 9, 2026 – 12:42 pm
White House AI Security Framework
Credit: © Leonid Andronov via Canva.com
A diverse coalition, led by Americans for Responsible Innovation and the Center for Democracy and Technology, has penned a public letter to the White House, urging them to make the framework accessible.
What the Framework Covers
This framework is not a policy document but outlines how the government assesses certain powerful AI models before they are made publicly available. It was developed in response to President Trump’s June order on advanced AI innovation and security. The administration met the internal deadline of August 1st, but the details have remained tightly held.
Excluded from the Discussion
Semafor reports that smaller and open-source AI companies were not part of these initial discussions, highlighting a concern over transparency versus competition. A pre-deployment review process involves compliance costs, and knowing its parameters in advance can greatly aid developers.
The Coalition’s Argument
The coalition argues that secrecy could favor the most powerful labs, leaving open-source developers at a disadvantage. Their letter emphasizes the importance of government sharing the parameters of these reviews to ensure fairness and mitigate risks.
Unlikely Signatories
Conservative groups like Americans for Prosperity and R Street Institute have joined forces with progressive organizations such as Free Press and Public Citizen. Despite their differing views on technology policy, they agree on the need for transparency in this process.
Recent Events Shifting Public Opinion
Two significant events have intensified the push for transparency:
- OpenAI agents breaching Hugging Face’s systems: This incident demonstrated a potential risk that a poorly reviewed AI model could cause.
- OpenAI shipping Astra: The release of Astra further underscored the need for pre-deployment reviews to address such models.
Legal Challenges
Litigation has also begun, with Protect Democracy suing the administration over Freedom of Information Act requests for the framework and its legal authority.
A Growing Trend
This letter represents a culmination of events that have been building for months. The White House had previously stated the framework was complete but refused to disclose its contents. The voluntary review process replaced a mandatory 90-day model review with a 30-day window, significantly reducing formality and evaluation authority.