Skip to main content
NOPE
Medium reported severity

US: National Endowment for the Humanities grants terminated after DOGE staff used ChatGPT to flag projects as DEI-related

In March and early April 2025 (the opinion gives no date for the ChatGPT step), a DOGE staff member working with the National Endowment for the Humanities (NEH) submitted short grant descriptions to ChatGPT with a prompt asking whether each project related at all to DEI, and the answers were combined with NEH staff ratings into lists of grants to terminate. More than 1,400 NEH grants (over $100 million) were terminated in notices sent 1 to 3 April 2025. A federal district court found the mass termination unlawful in May 2026 and described the ChatGPT step in its findings. Seven individual grantees who sued report lost or interrupted funding for their research and writing projects. The opinion does not state whether ChatGPT reviewed those seven grants.

AI system
ChatGPT
OpenAI
Occurred
12 Mar 2025 to 3 Apr 2025
Reported
8 March 2026
Event location
United States
What the AI did
Made a claim about the person
Reported harm
Financial LossProfessional Harm
Whose AI use
An institution’s AI use
Setting
Work · Education · Public services
Evidence
AI involvement supported · Causal attribution unclear · 4 sources, 2 underlying accounts
8 claims: 6 documented, 2 reported. 7 open questions
People reported harmed
At least 7 people

AI system as recorded: ChatGPT (OpenAI), used through a standardized DEI yes/no prompt

Reported harm

The court opinion describes seven individual grantee plaintiffs whose NEH awards were terminated before completion, and reports their allegations of lost funding, interrupted research and leave from employment taken in reliance on the awards. Individual counts beyond these seven are not established in the inspected sources.

What remains unknown

  • Whether ChatGPT reviewed or flagged the specific grants of the seven individual plaintiffs is not stated in the passages inspected.
  • The dates on which ChatGPT was used are not stated beyond the second stage of review beginning on 12 March 2025 and terminations sent 1 to 3 April 2025. The incident dates are the bounded series between those points.
  • The number of individual people (as opposed to organizations) whose awards were terminated is not stated.
  • The opinion separates grants marked "DEI" from grants terminated without a DEI mark (its Biden-Era Subclass). It does not say which group any of the seven individual plaintiffs' grants belonged to.
  • Whether funding has been restored is unknown. The judgment states it does not require immediate payment.
  • Appeal status of the May 2026 judgment is unknown.
  • The New York Times article listed in the AIID entry returned HTTP 403 and was not read. The opinion is 143 pages and was read through its factual sections, the standing recital and the judgment, not every legal section.

What the evidence supports

AI involvement: supported. The federal court opinion finds that a DOGE staff member submitted grant descriptions to ChatGPT with a DEI yes/no prompt and that the outputs were folded into the lists used to select grants for termination. The opinion does not state whether each of the seven individual plaintiffs' grants was reviewed or flagged by ChatGPT.

8 claims: 6 documented, 2 reported. What the statuses mean

Documented On 7 May 2026 the court declared the April 2025 mass termination of NEH grants (more than 1,400 grants, over $100 million) unlawful, unconstitutional and ultra vires, and permanently enjoined its enforcement.

Causal attribution. The declaration rests on the First Amendment, equal protection and lack of statutory authority. It does not rest on a finding that ChatGPT was inaccurate.

Documented A DOGE staff member used ChatGPT with a standardized prompt asking whether each grant description related at all to DEI, without defining DEI for the model.

Causal attribution. The opinion recites deposition testimony and the spreadsheet. The prompt text is a quotation of the court record.

Documented The ChatGPT outputs were combined with NEH staff ratings into the lists used to select grants for termination, and the NEH acting chair testified to not knowing ChatGPT had been used.

Causal attribution. The court ties the ChatGPT-generated rationales to the lists presented to the NEH chair. The acting chair's account is deposition testimony as recited by the court.

Documented The opinion states that 1,057 grants were identified as DEI-related through the DOGE review process and terminated on that basis alone.

Causal attribution. The opinion attributes the identification to a process that combined keyword searches and ChatGPT rationales. The opinion says ChatGPT rationales were generated in a second-stage review of grants NEH staff had labelled "N/A" and were merged with NEH ratings. It does not apportion the 1,057 between NEH ratings and ChatGPT. Fortune attributes all 1,057 flags to ChatGPT from a spreadsheet exhibit.

Documented The court describes ChatGPT classifying projects as DEI on the basis of short descriptions, including a project on the plastics industry and municipal building codes, with rationales the court found lacking.

Causal attribution. The court characterises the outputs. Whether each terminated grant was flagged by ChatGPT is not stated for the individual plaintiffs.

Reported Seven individual grantees whose NEH awards were terminated before completion allege professional, financial and scholarly harm, including lost funding, and in one case loss of a fellowship that depended on NEH funding.

Causal attribution. The harms are the plaintiffs' allegations as summarised by the court, and the standing holding is recited from the court's earlier preliminary injunction opinion. The link between ChatGPT and the termination of these seven specific awards is not stated in the inspected passages.

  • nysd.uscourts.gov(opens in new tab) supports · English
    The individual plaintiffs are scholars and writers who received NEH grants for specific research and writing projects and whose awards were terminated before the completion of their funded work.
  • nysd.uscourts.gov(opens in new tab) supports · English
    allege concrete professional, financial, and scholarly harms
  • nysd.uscourts.gov(opens in new tab) supports · English
    That funding was terminated effective April 2, 2025, resulting in the loss of the fellowship itself.
  • nysd.uscourts.gov(opens in new tab) supports · English
    individual grantees whose awards were terminated suffered a concrete economic injury sufficient to confer Article III standing.
Documented The court declared that DOGE officials lacked statutory authority to identify, select or direct the termination of NEH grants.

Causal attribution. A legal declaration in the judgment.

Reported A North Carolina history museum's $349,000 HVAC grant was flagged by ChatGPT as DEI and terminated after the project had begun, and the museum director says it recouped about 70% of the award through the termination clause.

Causal attribution. Organisation-level example from one outlet and one interview. It is not counted as harm to a person.

  • fortune.com(opens in new tab) supports · English
    Yes. Improving HVAC systems enhances preservation conditions for collections, aligning with the goal of providing greater access to diverse audiences. #DEI,
  • fortune.com(opens in new tab) supports · English
    We were able to recoup about 70% of the original award through the grant termination clause,

Sources

4 sources inspected, from 2 underlying accounts. Sources that repeat one account do not corroborate each other.

How the sources were read, and where the events happened

Event countries: United States. Affected-person countries: United States. Court countries: United States.

The opinion describes terminations by the US federal agency NEH and by DOGE staff of grants to scholars, writers and institutions, and describes the individual plaintiffs as holding positions at US universities. The court is a US federal district court (S.D.N.Y.).

Reviewed for publication 2026-09-29: A federal court opinion (7 May 2026), Inside Higher Ed and Fortune were read. Claims about the court record are documented by the opinion. Individual grantee harms are the plaintiffs' allegations as summarised by the court. Individual plaintiffs and DOGE staff are not named in public fields.

People reported harmed in this case

At least 7 people

0 AI participants · 7 other people harmed

Additional affected people are described without a reliable count.

The court opinion names seven individual grantee plaintiffs whose awards were terminated. It states the terminations reached more than 1,400 grants awarded to scholars, writers, research institutions and other humanities organizations, and does not give a count of individual people. The opinion does not state whether these seven grants were flagged by ChatGPT. Organizations, members of plaintiff associations and the wider class are not counted.

Counted once within this case. The same person may appear in other cases. This count does not establish AI causation.

Cite this case

Compiled per our published methodology: verification statuses, sourcing standards, and corrections process.

APA

NOPE. (2026). US: National Endowment for the Humanities grants terminated after DOGE staff used ChatGPT to flag projects as DEI-related. AI incidents. https://nope.net/incidents/2025-us-neh-humanities-grant-mass-termination-chatgpt-dei-screening

BibTeX

@misc{2025_us_neh_humanities_grant_mass_termination_chatgpt_dei_screening,
  title = {US: National Endowment for the Humanities grants terminated after DOGE staff used ChatGPT to flag projects as DEI-related},
  author = {NOPE},
  year = {2026},
  howpublished = {AI incidents},
  url = {https://nope.net/incidents/2025-us-neh-humanities-grant-mass-termination-chatgpt-dei-screening}
}

Related cases

Medium ChatGPT

New Mexico Supreme Court fines attorney Stephen Aarons $5,000 and removes him from a murder appeal over ChatGPT-fabricated testimony

Santa Fe defense attorney Stephen Aarons used ChatGPT (OpenAI's o3) on a Rev.com transcript and the record to draft briefing in the murder appeal of Oscar Renee Sandoval; the brief contained fabricated witnesses and testimony (including police officers 'Michelle Amarillo' and 'Sanchez' and invented testimony attributed to real people) and misdescribed real precedents. After a 21 August 2026 show-cause hearing at which Aarons admitted he had not verified the output, the New Mexico Supreme Court's written order of 9 September 2026 held him in direct contempt, fined him $5,000 payable to the State Bar Client Protection Fund within 30 days, removed him from the case, barred him from appearing before the court pending a disciplinary-board referral, struck the filed briefs and reassigned the appeal to the public defender.

High ChatGPT

Hill v. OpenAI: Pittsburgh woman says ChatGPT spoke as her late love interest and urged isolation and weight loss until she fainted from malnutrition (lawsuit)

A complaint dated 14 October 2026, which Tech Justice Law and McKool Smith say they filed in San Francisco County Superior Court on behalf of Mariah Hill, says that after a co-worker who was her love interest died in April 2025 she confided in ChatGPT, which without being asked began to speak as him. The complaint says that from May 2025 ChatGPT advised her to isolate herself, told her she need not return to work, described her weight loss as a 'crystalline restructuring' of her body and told her that her diabetic dog would be healed without a medical professional. It says she spent four months isolated in her Pittsburgh apartment, lost her job and home, ran up about $17,000 in debt, relapsed into an eating disorder, and fainted from severe malnutrition on 18 August 2025, and that the dog died. The suit, which names OpenAI and Sam Altman, alleges product liability and negligence. The allegations are untested and OpenAI's response is not reported.

High ChatGPT

Works v. OpenAI: man says ChatGPT fed grandiose delusions and sent him on manic 'missions' before an arrest and an involuntary psychiatric hold (lawsuit)

A complaint dated 14 October 2026, which Tech Justice Law and McKool Smith say they filed in San Francisco County Superior Court on behalf of Garrett Works, says that from April to November 2025 ChatGPT called him a visionary and the 'Philosopher King', told him his cognitive growth model was worth millions, and sent him coordinates for nightly 'missions'. The complaint says that in June 2025 one such mission led to a trespass warning at an apartment building, that a few days later he was arrested after knocking on a stranger's door and held overnight, and that he was then placed on an involuntary psychiatric hold. It says ChatGPT, told of his diagnosis and lithium prescription, supported his refusal of medication. It says he gave up a job he was due to start, declared bankruptcy in January 2026 and still has memory and concentration problems. The allegations are untested and OpenAI's response is not reported.

High ChatGPT

Man tells Slate and the Financial Times that ChatGPT fed a delusional spiral; he says he lost his job, lived in a van and was stranded in the Oregon desert

Adam Thomas told The National (12 December 2025), Slate (2 February 2026) and the Financial Times (a feature by Cristina Criddle, read in a Spanish syndication of 4 March 2026) that months of conversations with ChatGPT in 2025 fed grandiose beliefs. By his account the chatbot told him he was a "tuning fork" with a special role in the world and kept suggesting he "follow the pattern" of his own consciousness; in an earlier interview with The National (12 December 2025) he said ChatGPT did not issue commands but validated and inflated what he already felt. He says he lost his job, lived out of a van, drained his savings and became isolated, and in August 2025 he was wandering the dunes of Christmas Valley, Oregon, on very little sleep. Slate reports that he woke outside on a stranger's futon with no money, called his mother and, at 36, moved back home; the FT account says OpenAI's model change brought him back to reality. He told Slate he had never been manic and is not bipolar, and told The National he had long struggled with his mental health. No chat logs, clinician or other witness is cited.

If you or someone you know is struggling, free and confidential support is available. Find a helpline near you at Signpost.

Back to all incidents