Protect Democracy sues agencies over secret US AI safety test rules

In short: Protect Democracy filed a lawsuit to force four federal agencies to release unclassified details about the Trump administration’s secret process for reviewing advanced AI models before they are released.

What happened

Protect Democracy, a nonpartisan nonprofit, sued four parts of the federal government, the Office of the National Cyber Director, the Office of Science and Technology Policy, the Treasury Department, and the Commerce Department. The group says the public and Congress have been given “almost no details” about the framework the government uses to review “frontier” AI models, meaning the most capable and potentially risky models.

The lawsuit asks a court to order the agencies to produce documents by September 30. Protect Democracy says it is only seeking unclassified information, like the framework’s text, who is participating, what the rules are for participating, and how the government decides who can access these powerful models. It also wants the court to stop agencies from withholding records that are not classified.

Protect Democracy argues that the secrecy could hide favoritism or retaliation. It pointed to a recent judge’s ruling that it was illegal for the Trump administration to blacklist Anthropic. The group also raised concerns about “GOLD EAGLE,” a White House program that relies on unnamed industry partners to help share and coordinate information about cybersecurity weaknesses (like a neighborhood watch for software problems).

Why it matters

These reviews can affect which AI products reach the public, and who is allowed to use the most capable systems. If the rules are secret, it is harder for Congress and the public to judge whether the tests are strong, fair, and up to date, or whether they are being used to reward allies and punish critics.

Source: Arstechnica