Protect Democracy sues to reveal Trump's secret AI safety framework

The lawsuit comes amid rising concerns about the executive branch's expanding role in AI governance. The White House set up the voluntary framework after flagging Anthropic's Mythos 5 model as too dangerous, and has since confirmed both the framework and GOLD EAGLE are in use. Protect Democracy argues that without public disclosure, the administration could use the process to coerce AI companies on issues like lethal autonomous weapons and mass surveillance, or to extract political support from investors and employees.

The case highlights a broader transparency gap: while some parts of the process are classified, the framework itself is not, yet the White House has refused to share details, with a spokesperson saying that unclassified information may still be withheld from public broadcast. The complaint also notes that the term 'covered frontier model' lacks a clear definition, which could either let dangerous models slip through or overburden agencies already affected by DOGE cuts. Meanwhile, Congress faces a decision on extending CISA provisions, which a senior official said is fundamental to GOLD EAGLE's viability.

Protect Democracy's action is part of its broader mission to prevent democratic decline, and it has previously used FOIA requests to seek transparency. The group says it will publish all records it obtains, aiming to allow public and congressional oversight of a process that could shape the future of AI development and national security.