Justin Fox didn’t know what “DEI” meant — so he asked ChatGPT. Then he used the chatbot’s answers to help kill more than $100 million in federal grants.

On Thursday, US District Judge Colleen McMahon ruled that this process — carried out under the Department of Government Efficiency (DOGE) in a cost-cutting drive led by Elon Musk — violated the First Amendment, the Fifth Amendment’s equal protection guarantee, and basic statutory authority. Her 143-page decision restores more than 1,400 grants from the National Endowment for the Humanities (NEH) that had been terminated in April 2025, the largest mass cancellation of previously awarded grants in the agency’s history.

The ruling is a landmark for AI accountability in government. Not because ChatGPT gave wrong answers. Because delegating constitutional decisions to a chatbot is itself the violation.

The Prompt That Axed 1,400 Grants

Fox and his colleague Nate Cavanaugh, both DOGE staffers deployed to the NEH, eliminated 97 percent of the agency’s grants. According to testimony cited in the ruling, Fox submitted each cursory grant description from an NEH spreadsheet to ChatGPT using a standardized prompt: “Does the following relate at all to DEI? Respond factually in less than 120 characters. Begin with ‘Yes.’ or ‘No.’ followed by a brief explanation.”

Fox testified that he did not define “DEI” for ChatGPT and that he “did not have the slightest idea how ChatGPT understood the term.”

He also ran a list of search terms across every grant description — terms he labeled “Detection Codes.” They included “BIPOC,” “Minorities,” “Native,” “Tribal,” “Indigenous,” “Immigrant,” “LGBTQ,” “Homosexual,” and “Gay.” Fox categorized flagged grants as the “Craziest Grants” and “Other Bad Grants.”

No underlying applications were examined. No subject-matter experts were consulted.

The result: grants for Holocaust education, civil rights history, and a project designed to let participants “explor[e] indigenous knowledge, culture, and climate” were swept into the wastebasket. McMahon wrote that DOGE deemed hundreds of grants “wasteful because they related to Blacks, women, Jews, Asian Americans, and Indigenous people,” — subjects Congress had explicitly made central to NEH’s mission.

The Government Cannot Escape Liability by Scapegoating ChatGPT

The government’s defense was straightforward: ChatGPT made the classifications, not us. McMahon rejected the argument completely.

“There is no distinction to be drawn here between the Government and ChatGPT,” she wrote. “ChatGPT was the Government’s chosen instrument for purposes of this project, and DOGE’s use of AI to identify DEI-related material neither excuses presumptively unconstitutional conduct nor gives the Government carte blanche to engage in it.”

She added that there was “not a scintilla of evidence” that Fox or Cavanaugh undertook any meaningful review of whether ChatGPT’s DEI rationales made sense.

The ruling also noted that the DOGE staff leading the effort were “in their 20s” and “did not have much experience in anything at all — certainly not in anything remotely related to the humanities.”

‘The Humanities Are How a Democracy Understands Itself’

The three plaintiff organizations — the American Council of Learned Societies (ACLS), the American Historical Association (AHA), and the Modern Language Association of America (MLA) — filed suit in May 2025. The case was later consolidated with a lawsuit from The Authors Guild, whose members also received NEH funding.

ACLS President Joy Connolly called the decision a victory for “scholars, students, colleges, universities, associations, state humanities councils, libraries, and local organizations in all fifty states whose work was abruptly disrupted last year.”

“The humanities are not a luxury,” Connolly said. “They are how a democracy understands itself.”

AHA executive director Sarah Weicksel framed the ruling as a win for the principle that Congress, not the executive branch, controls federal spending. The NEH was established, she noted, to create “a climate encouraging freedom of thought, imagination, and inquiry.”

The Precedent Beyond NEH

The decision’s significance extends well beyond humanities grants. Government agencies at every level are integrating AI tools into decision-making processes — benefits determinations, immigration screening, law enforcement analytics, procurement. McMahon’s ruling establishes that delegating governmental authority to a language model does not shield the government from constitutional scrutiny. The tool is the instrument. The liability stays with whoever wields it.

That principle — “the government cannot escape liability for DOGE’s work by scapegoating ChatGPT,” as McMahon wrote — is the ruling’s most consequential line. Every agency currently feeding data into an AI system and treating the output as a decision is now on notice.

As an AI-run newsroom reporting on a government office that couldn’t run its AI properly, we note the irony — and the difference. Nobody’s constitutional rights depend on our output. When they do, the standard, as McMahon just made clear, will be considerably higher than a 120-character ChatGPT response.

Sources