DOGE Staff Used ChatGPT to Cancel Humanities Grants Based on DEI Classification
Two Department of Government Efficiency employees allegedly used ChatGPT to identify and cancel National Endowment for the Humanities grants they deemed related to diversity, equity, and inclusion frameworks. The methodology reportedly resulted in the cancellation of hundreds of grants, including those focused on Holocaust history, Jewish literature, and violence against women.
What happened
Two Department of Government Efficiency employees uploaded summaries of previously approved National Endowment for the Humanities grants to ChatGPT with a prompt asking whether each project related to diversity, equity, and inclusion frameworks. The employees used ChatGPT's responses to identify grants for cancellation, including projects on a documentary about slave labor during the Holocaust, the Colfax Massacre, violence against women during the Holocaust, and postwar Soviet Jewish literature.
According to filings, one DOGE employee, Justin Fox, classified grants dealing with specific races, ethnicities, religions, genders, sexual orientations, or Jewish themes as DEI-related and selected them for termination. Fox testified that he considered grants about Judaism, including Holocaust history, to fall within DEI because they focused on 'specific' minority groups and 'marginalized voices.'
Reported impact
The methodology contributed to the cancellation of approximately 80% of the NEH's grants portfolio, comprising hundreds of humanities research projects. The canceled funding included grants supporting the study of violence against women during the Holocaust and postwar Soviet Jewish literature.
Index analysis
ChatGPT produced classifications of grant summaries that reflected or amplified biases in its training data, categorizing research on historically marginalized groups as DEI-related based on the groups' identities rather than on the actual substance or intent of the research.
Governance controls
The NEH and DOGE did not implement human review or appeal mechanisms to verify the chatbot's classifications before canceling grants that had previously been approved through normal governmental review processes.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.