What happened

U.S. Department of Government Efficient (DOGE) operatives used ChatGPT to conduct an "audit" of the National Endowment for the Humanities (NEH), feeding the summaries of thousands of active grants into the AI with a prompt asking for a "Yes/No" response on whether the projects related to Diversity, Equity, and Inclusion (DEI).

Based on these 120-character AI outputs, the NEH issued immediate termination notices to hundreds of institutions, resulting in some institutions facing bankruptcy, stopping projects, and firing staff.

Over USD 100 million in funding was clawed back by DOGE.

Targeted projects included a documentary on Jewish slave labour during the Holocaust, the digitisation of Appalachian historical records, and archives of Indigenous languages.

The NEH itself saw roughly 70 percent of its staff placed on administrative leave or fired.

Why it happened

The incident was driven by a desire for rapid, "revolutionary" cost-cutting that bypassed traditional bureaucratic and scholarly review.

Rather than experts reviewing the grants, DOGE relied on a Large Language Model (LLM) to "tag" complex historical and cultural research based on a narrow list of keywords like "BIPOC," "LGBTQ," and "gender."

DOGE operated as an "outside" advisory body, allowing its staff to bypass federal hiring rules and conflict-of-interest laws.

Discovery in a 2026 lawsuit revealed that DOGE and NEH staff used Signal, with auto-delete enabled, to communicate, a direct violation of the Federal Records Act intended to hide the decision-making process.