Court Filings Show DOGE Used ChatGPT to Flag More Than 1,000 Humanities Grants as 'DEI'

In a brief filed March 6, 2026, in the Authors Guild's class action over National Endowment for the Humanities grant terminations, plaintiffs laid out discovery showing that DOGE staff used ChatGPT to decide which grants to cancel. The prompt they used asked the model: does the following relate at all to DEI, respond factually in less than 120 characters, beginning with yes or no. Staff supplied no definition of DEI and accepted the answers without review. More than 1,000 grants were initially selected on that basis, and the sweep later extended to remaining Biden-era awards; about 1,400 grants were terminated in total, roughly 97 percent of the agency's grants. Reported examples include a $349,000 grant to replace a museum's HVAC system, flagged as aligning with DEI goals of inclusivity, and a newspaper digitization project flagged for promoting accessibility. GSA and DOGE staffer Justin Fox applied the screen; a DOGE supervisor told NEH staff they were under pressure from the White House.

The filings describe an automated tool making final terminations that Congress assigned to the agency, without a governing standard or human check.

The court had already entered a preliminary injunction finding First Amendment and Administrative Procedure Act violations. On May 8, 2026, a judge ruled the cancellations unconstitutional.