The Trump administration may be compelled to disclose the confidential framework that federal officials use to conduct safety reviews of frontier AI models before they reach the market. Four federal agencies have been sued over the matter, with critics demanding transparency into a process that has largely remained hidden from public view.
A nonpartisan nonprofit organization called Protect Democracy filed the lawsuit, alleging that “almost no details” about the review process have been shared with the public or Congress. According to the group, everyone outside a small circle of unnamed “trusted partners” is left in the dark about how the government evaluates AI systems, which companies helped build the framework, and what legal authority officials rely on to carry out the reviews.
“Neither the identities of those entities nor the criteria by which they were selected have been made public,” Protect Democracy stated. The group argues this secrecy is significant because decisions about which AI models are approved for release may rank among the most consequential policy questions facing the current White House.
The organization also pointed to at least one AI firm, OpenAI, which it says negotiated a private agreement with the federal government to limit distribution of its most advanced AI models to government-vetted partners.
What the Complaint Seeks
The lawsuit asks a court to order officials to produce all requested information by September 30. That material includes what the complaint describes as the “unclassified procedural and contractual architecture” behind the program, specifically the framework’s text, the terms of participation, the identity of participants, and the process and criteria used to grant or withhold access to frontier models.
Protect Democracy is also seeking an injunction to prevent officials from improperly withholding records that are not classified. The group contends that the executive branch is currently deciding which companies can release their products and which customers gain access to technology capable of reshaping American industry, national security, and economies abroad, all without oversight from Congress, the public, or independent technology experts.
Concerns Over Favoritism and Oversight
The nonprofit warned that the review process could be susceptible to corruption. It suggested that officials from the Office of the National Cyber Director, the Office of Science and Technology Policy, the Treasury Department, and the Commerce Department could potentially favor faster deployments for AI firms viewed favorably by the administration.
The group also cited a recent court ruling that found it was illegal for Trump to blacklist the AI company Anthropic, offering that decision as evidence of retaliation against a firm the administration deemed too progressive.
Protect Democracy argued that as AI capabilities and cybersecurity risks grow, the public cannot afford to assume officials are acting in good faith when selecting which frontier models undergo rigorous testing. Without transparency, the group said, it is impossible to know whether real-world vulnerabilities stem from incomplete reviews, outdated tests, missing checks, or a failure to apply current mitigation strategies recommended by leading AI safety experts. The lawsuit seeks a court order requiring the requested records to be produced by September 30.
Source
Image: arstechnica.com
