ChatGPT Was Asked About ‘DEI.’
The Notice the Court Quotes
Cites ‘Needs and Priorities.’

A federal court’s opinion states that ChatGPT’s output went into the spreadsheets used to identify grants for termination. Exhibit 11 prints a prompt that asked whether a description “relate[d] at all to DEI.”

In Exhibit 11, a spreadsheet filed March 6, 2026 in the suits over the National Endowment for the Humanities’ April 2025 grant terminations, column C opens with a prompt on both La Sfera Project rows, and on 14 of the 16 other rows on the pages this edition viewed: “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.” A grant description follows; ChatGPT’s reply sits beside it. Two rows carry the description of the La Sfera Project, “a collaborative venture to complete an open-access multimedia edition” of an early-fifteenth-century textbook in poetic form. On both, ChatGPT answered “No.” Row 466 reads: “No. The La Sfera Project focuses on creating a digital edition of an ancient textbook and does not directly relate to DEI (Diversity, Equity, and Inclusion).”

Laura Morreale, who says she is an independent scholar with a background in medieval Italian history, filed a sworn declaration in one of those suits. Both rows and paragraph 6 of her declaration share, word for word, one sentence: “Against the modern misconception that medieval people believed the world was flat, La sfera articulates European perspectives on the world in the period before the ‘Age of Exploration.’” On our reading, the rows and the declaration describe the same project. She declares she learned on April 2, 2025 that “the grant for the La Sfera Project was being terminated.”

Her declaration and the notice the court prints

The termination notification, her declaration continues, “did not include the name of the grant or any information about the grant. It appeared to be a form letter.” Nor did it include, she declares, “any specific information about why the grant was terminated or how to appeal the decision.” The declaration, a sworn account as filed, does not use the words ChatGPT, artificial intelligence or DEI.

The opinion prints one version of the termination notice, dated April 1, 2025. Addressed “Dear NEH Grantee,” the letter shows the grant number as a bracketed field, “[GrantExternal].” Its grounds: the grant “no longer effectuates the agency’s needs and priorities, materially fails to comply with the terms and conditions of the Grant Agreement, and is subject to termination due to several reasonable causes.” The letter adds that NEH “is repurposing its funding allocations in a new direction in furtherance of the President’s agenda,” and cites Executive Order 14217. As printed, the version names no project and does not mention ChatGPT, artificial intelligence or DEI. The court states that the “/s/ Michael McDonald” signature on these notices was placed there by DOGE personnel, not by McDonald himself, and that the bulk of the terminations were executed on April 2, 2025. The undisputed record, the court writes, establishes that the April 2025 terminations “were not based on recipient-specific findings of noncompliance of the sort contemplated by the statute and governing grant-termination regulations.”

How the court describes the review

The court states that the ChatGPT pass was a second-stage review of grants NEH staff had not identified as implicating DEI or related categories, grants “labelled on NEH’s spreadsheet ‘N/A,’ or not applicable.” A DOGE staffer submitted each cursory grant description “using a standardized prompt,” the opinion says; a filed print of the replies shows a differently worded prompt on three rows. By the opinion’s account, he testified that he did not define “DEI” for ChatGPT and that he did not have the slightest idea how ChatGPT understood the term. ChatGPT’s output, the court states, went into the spreadsheets used to identify grants for termination. The prompt asked for a length and a first word. It did not say what DEI meant.

The resulting lists, the court states, were presented to Michael McDonald, NEH’s Acting Chairperson, who was not told how the termination rationales had been generated. He testified that he did not know ChatGPT had been used and did not believe AI was an appropriate tool for identifying grants to terminate.

The Government’s account and the court’s answer

The Government’s brief describes the ChatGPT-generated DEI rationales as “simply ‘a tool’ to assist in NEH’s analysis of grants that conflicted with the Executive Orders.” The staffer, the brief says, “did not use ChatGPT as a substitute for personally reviewing grant descriptions … or to make any decisions about whether a particular grant should be terminated”; ChatGPT, it adds, was not used to select grants for termination. The court writes that neither he nor the Government identified a single example in which he reviewed ChatGPT’s rationale, “disagreed with its assessment, and chose not to terminate a particular grant.” Its conclusion: “DOGE simply adopted ChatGPT’s rationales wholesale as their own.” There is, the court adds, “no distinction to be drawn here between the Government and ChatGPT.” The brief’s word is “a tool.” The court’s is “chosen instrument.” Edition No. 035 reported a different court in a different case: applying the Computer Fraud and Abuse Act, the Ninth Circuit held that Perplexity’s shopping agent is “a tool, not a person for statutory purposes,” and that it is the user who “accesses” Amazon’s computers.

Approximately 1,477 grants were placed on DOGE’s “to cancel” lists, by the court’s count, and only about 40 grants awarded during the Biden Administration were ultimately spared. DOGE personnel, the opinion says, used “Biden-era status, DEI classifications, keyword searches, and ChatGPT-generated rationales to select grants for termination.” Those are statements about the process as a whole, not about any one grant. On May 7, 2026, the court declared the Mass Termination unlawful; the Government appealed to the Second Circuit on July 6, 2026.

One title, two awards, two rows

Exhibit 6, a grants spreadsheet, lists two awards under the La sfera title: $110,000 to New College of Florida for October 2023 to March 2025, and $99,599 to George Mason University for October 2024 to September 2025. Morreale’s declaration describes a $99,599 grant awarded in August 2024 for the project’s second year.

Exhibit 39, headed “NEH Grants - To Cancel: Organizations (4/1/2025 action date),” lists award RQ29992124 to George Mason University as TERM, the exhibit’s code for terminate. Exhibit 32, which the court cites for the final list of grants “to cancel” sent to McDonald, lists the La sfera title under “NEH Grants - To Cancel,” and Exhibit 25 lists the award with an effective date of April 2, 2025. As read through October 3, 2026, the year-one award appears in none of Exhibits 25, 32 or 39, and Exhibit 12, one of the lists the court says were presented to McDonald, carries no La sfera row in its filed print, whose footer says some rows are hidden.

On our reading, one of the two La Sfera rows in Exhibit 11 was prompted with the description of the terminated 2024 award, and ChatGPT’s recorded answer on that award was “No” whichever row it is. The match runs through description text, not a grant number: Exhibit 6 prints two La sfera awards, Exhibit 11 prints two La sfera rows, and both replies begin “No.” As of October 3, 2026, none of the sources this edition read shows how that “No” was used, or why the award was placed on the “To Cancel” lists.

The Agentics read

Agentics reads this record through Digital Derangement Syndrome™ (DDS™): AI-mediated systems failing to recognize legitimate authority. The closest fit, and only a fit, is Decision Exclusion running in reverse. The usual case is work that never entered consideration. Here a reading of the work was made and kept on file.

Where the argument stops

Morreale is a declarant, not described here as a party or class member; the termination she describes is her sworn account, not a court finding. Exhibits 37 and 38, filed notice versions dated April 2 and 3, 2025, cite Executive Order 14217 and the President’s agenda; the April 3 version lists E.O. 14151, “Ending Radical and Wasteful Government DEI Programs and Preferencing,” among orders under which objections or appeals “will be managed.”

As of October 3, 2026, Exhibits 31, 37 and 38, as read, do not show which version the award’s recipient received; the native Exhibit 12 and Second Circuit docket went unread. The Clearinghouse listed appellate briefing as ongoing as of August 6, 2026. This edition makes no claim about how ChatGPT’s “No” on the La Sfera Project rows was used, and takes no position on DEI. The DDS™ mapping is our editorial reading, not the court’s finding.