DOGE Uses ChatGPT to Flag and Cancel Federal Humanities Grants (2025)

Incident Summary

In April 2025, the Department of Government Efficiency (DOGE) deployed OpenAI’s ChatGPT to screen National Endowment for the Humanities (NEH) grant descriptions, instructing the model to flag grants as “DEI” with a yes/no classification and accompanying rationale.^[1] The ChatGPT-generated list replaced a prior assessment compiled by NEH subject-matter staff, and NEH subsequently eliminated the flagged grants, associated grant programs, staff positions, and entire divisions.^[2]

The scope of disruption exceeded $100 million in humanities projects. Court discovery documents released on March 6, 2026, in US District Court for the Southern District of New York revealed the mechanism after the American Council of Learned Societies (ACLS), American Historical Association (AHA), and Modern Language Association (MLA) filed suit on May 1, 2025.^[1]

Key Facts

• Method: ChatGPT was prompted to classify NEH grant descriptions as DEI-related with binary yes/no and a rationale^[1]
• Replacement of expertise: The AI-generated list replaced an assessment prepared by NEH professional staff^[2]
• Terminated projects included: A documentary on Holocaust-era Jewish women’s slave labor, an Italian American archival project, Appalachian photograph digitization, and Native American language preservation programs^[3]
• Misclassification: An HVAC infrastructure grant for a museum was flagged as DEI-related^[4]
• Scale: Over $100 million in humanities projects disrupted^[1]
• Legal action: ACLS, AHA, and MLA filed lawsuit May 1, 2025; discovery released March 6, 2026^[2]

Threat Patterns Involved

Primary: Proxy Discrimination — ChatGPT’s classification of grants as “DEI” operated as a proxy filter that disproportionately flagged projects involving minority communities, cultural heritage preservation, and historically marginalized groups, without evaluating the substantive merit or content of the grant work itself.

Secondary: Overreliance & Automation Bias — DOGE operators substituted ChatGPT’s classification for expert human judgment by NEH staff, treating the model’s binary output as an authoritative determination suitable for consequential funding decisions affecting hundreds of millions of dollars in public humanities investment.

Significance

This incident represents one of the first confirmed cases of a government body using a commercial large language model to make consequential policy decisions affecting public funding at scale. The case is significant for several reasons:

1. Substitution of expertise — AI classification replaced domain-expert assessment for decisions with irreversible cultural consequences
2. Proxy discrimination at institutional scale — The use of “DEI” as a binary classification category resulted in the termination of projects spanning Holocaust documentation, Indigenous language preservation, and regional cultural heritage, none of which are inherently DEI programs
3. Accountability gap — The mechanism was only revealed through litigation discovery, raising questions about transparency in government AI deployment
4. Precedent-setting litigation — The pending lawsuit may establish legal standards for governmental use of AI in funding and policy determinations