A federal judge in New York ruled that the Trump administration’s cancellation of more than $100 million in NEH grants was unconstitutional, and found DOGE and the administration lacked authority to terminate the grants.

Testimony showed DOGE staff used ChatGPT to scan grant descriptions for DEI connections and to apply detection codes tied to protected characteristics to revoke funding.

Judge Colleen McMahon said the government violated First Amendment and equal protection by terminating grants based on perceived viewpoints and DEI classifications, calling it a textbook example of unconstitutional viewpoint discrimination.

The ruling underscores opportunities for AI infrastructure that is auditable, bias-tested, mapped to policies, and includes human-in-the-loop review to meet legal and procedural standards.

It demonstrates that government automation cannot bypass accountability, and officials remain responsible for final funding decisions even when aided by AI.

The decision cited specific DEI labeling of material, including an anthology on Jewish writers, illustrating AI-driven classification used to target funding.

Rights advocates warn the ruling could threaten decades of social progress and recognition of diverse histories.

The ruling may set a practical precedent for AI-assisted government action challenges, including demands for prompts, classification logic, statutory checks, and due-process proof.

The judge noted that a new administration can set lawful priorities but cannot suppress disfavored ideas, emphasizing the public interest in constitutional and congressional bounds.

The court stressed that public officials must act within the bounds set by Congress and the Constitution, even as new administrations pursue policy priorities.

Unclear whether the administration will appeal; White House and DOJ did not immediately respond to requests for comment.

The court highlighted the use of ChatGPT to screen grant descriptions for DEI links, with prompts favoring rapid, keyword-based decisions over defensible policy and human oversight.