DOGE used ChatGPT to gut federal grants

Using an LLM as a compliance oracle for billion-dollar policy decisions — with no human review of the outputs — is exactly the kind of AI misuse that makes the broader ecosystem look bad.

• –ChatGPT's responses were entered directly into a spreadsheet and used to override lists prepared by actual NEH subject-matter experts
• –The process flagged Jewish Holocaust documentation grants as "DEI," exposing how easily prompt-driven classification fails with culturally nuanced content
• –Neither staffer could define "DEI" under deposition, undermining any claim the AI was operationalizing a coherent policy standard
• –The lawsuit (ACLS, AHA, MLA v. NEH) raises First Amendment, Equal Protection, and separation of powers claims — the AI angle adds an accountability gap: who is responsible when a chatbot makes the call?
• –This will likely become a reference case in debates over AI in government decision-making and the liability gap when LLM outputs drive real-world harm