Court Filing Reveals AI and Keyword Searches Drove Federal Grant Terminations

New details have emerged regarding the termination of federal grants at the National Endowment for the Humanities (NEH). A proposed amended complaint filed by the Authors Guild against the U.S. government reveals that appointees Nate Cavanaugh and Justin Fox utilized ChatGPT to determine which grants to cut, often bypassing standard review protocols and expert staff.

According to the complaint, the review process lacked traditional oversight. Instead, Fox reportedly fed grant descriptions into ChatGPT with a specific command: "Does the following relate at all to DEI? Respond factually in less than 120 characters." There is no indication that the appointees verified the AI's interpretation of "DEI" or ensured it aligned with their own criteria. Consequently, grants flagged by this method were slated for termination, with NEH staff blocked from challenging the decisions.

Keyword Lists and AI Flags

Beyond AI queries, the decision-making process involved a "Detection List" of keywords. Reports indicate this list included terms such as "LGBTQ," "tribal," "immigrants," and "BIPOC." Conversely, terms like "white" or "heterosexual" were notably absent. Grants were categorized into spreadsheets labeled "Craziest Grants" and "Other Bad Grants," targeting projects focused on minority experiences and historical events.

The AI-driven searches captured a broad range of historical and cultural projects. Flagged grants included studies on the Colfax massacre, Jewish women’s slave labor during the Holocaust, and the history of Native American boarding schools.

Administrative Fallout

The complaint further alleges that termination emails were sent to over 1,400 recipients from a private, non-government email server. While these notices claimed authorization from acting NEH director Michael McDonald, he reportedly admitted he had no role in drafting the correspondence or selecting the grants for termination. McDonald stated that the methodology mirrored operations at other agencies rather than adhering to NEH's traditional processes.

The disclosures have raised significant concerns regarding the role of artificial intelligence in administrative decision-making and the displacement of expert review in federal grant allocations.