Labor & Employment Law Daily Wrap Up, LITIGATION NEWS, TRENDS—Court rules against DOGE regarding NEH grant terminations, (May 11, 2026)

By Brandi O. Brown, J.D.

“If strict scrutiny applies with such force to race-conscious districting undertaken in purported compliance with the Voting Rights Act... then it must apply with at least equal if not more force here, where race was used to exclude disfavored scholarship about historically marginalized groups from public support.”

In a decision issued late last week, a federal district court in New York examined the legality of the mass termination, by the National Endowment for the Humanities (NEH), of more than 1,400 grants in April 2025, a move that abruptly halted over $100 million in previously awarded funding for scholars, writers, research institutions, and other humanities organizations. Judge McMahon’s opinion presents a detailed account of how this action demonstrated a sharp break with the NEH’s statutory framework, its established grantmaking practices, and constitutional constraints. The opinion further offers insight into how the use of AI for decision making helped create legal problems for the federal government.

The NEH, created by Congress to support research, education, and public programming in the humanities, operates through a structured process involving peer review, advisory input, and final approval by the agency’s chair. Once grants are awarded, they have historically been terminated rarely and only in certain circumstances. Congress identified particular post-award consequences the chairperson could impose if the recipient substantially fails to satisfy the purposes for which the assistance was provided, including that the chairperson could take the results of the post-award evaluation into account in deciding whether to provide future assistance, could prohibit the recipient from using the Endowment’s name in connection with the project, and, if the project is published, could require a disclaimer that the views expressed do not reflect those of NEH.

The court explained that the provisions set out by Congress “show that Congress addressed post-award compliance in grantee-specific, Chair-centered terms, tied to compliance with award purposes, statutory criteria, and grant terms—not to a new administration’s ideological disagreement with previously awarded projects.”

Mass terminations. However, in early 2025 a series of executive actions aimed at eliminating federal support for projects associated with DEI and related concepts coincided with the formation of the Department of Government Efficiency (DOGE), which in turn became involved in reviewing NEH grants. This led to the mass termination of more than 1,400 grants, which halted over $100 million in previously awarded funding.

For example, one DOGE employee reviewed NEH grants using keyword searches of their descriptions, using terms such as “gay,” “BIPOC,” “indigenous,” “tribal,” “melting pot,” and “equality.” He created his own spreadsheet, which sorted all NEH grants from the Biden years into categories called “Craziest Grants” and “Other Bad Grants.” He later admitted that those labels were based on his and the other assigned DOGE employee’s “subjective” views and nothing indicated that either employee “read NEH’s authorizing statute or understood that Congress itself had identified many of the subjects swept up by those search terms as proper objects of NEH support.”

“Craziest grants” identified. By way of illustration, one of the “Craziest Grants” identified was a joint project between the Huntington Library and Art Gallery and the University of Southern California, which was designed to encourage public engagement with themes including “national identities,” “the struggle for democracy” from the perspective of California and the West, and “the nation’s uneven journey to an equitable and just society.” The period to be studied ran from 1450 to 1850, when California became a state.

AI is used. As part of the process, as it were, used by the DOGE workers to identify grants to be slashed, they also used ChatGPT. The DOGE employee used the artificial intelligence tool to create what the parties call “DEI rationales” that might arguably pertain to the grants, even though he testified that he neither defined “DEI” for ChatGPT nor had the slightest idea how the tool understood that term.

DOGE makes decisions. The DOGE workers ultimately presented their list to the Acting Chairperson of NEH. That official was not told how the termination rationales had been generated, and he later acknowledged that in some instances the stated rationales were insufficient, irrational, or nonsensical. While official tried to save some of the projects from the list, he acknowledged to the DOGE employees, “Either way, as you’ve made clear, it’s your decision on whether to discontinue funding any of the projects.” In other words, the court explained, the NEH acting chair recognized that he had no meaningful control over the grants’ fate and that the ultimate decision whether to continue or terminate funding rested not with NEH, but with DOGE. Indeed, his attempts to salvage some of the grants were unsuccessful.

Lawsuits filed. Two lawsuits were filed to challenge these actions, and they were consolidated. In one, the organizational plaintiffs are the American Council of Learned Societies (ACLS), the American Historical Association (AHA), and the Modern Language Association of America (MLA). In the other, the plaintiffs are The Authors Guild and individual NEH grantees who sued on behalf of themselves and a proposed class of similarly situated grantees.

Thet argue that the defendants—the NEH and its acting chairman, DOGE and DOGE officials, and the GSA—engaged in arbitrary and capricious agency action by terminating grants en masse without individualized consideration; that they acted without statutory authorization and effectively displaced the National Endowment for the Humanities Chair and the grantmaking process Congress prescribed by effectuating terminations outside the statutory scheme set forth in 20 U.S.C. §§ 956(c), 957(f); and that they discriminated on the basis of viewpoint and protected characteristics by targeting projects associated with “diversity, equity, and inclusion,” “gender ideology,” and other subject matter that the Trump Administration regarded as inconsistent with its priorities.

Ultra vires. In its consideration of cross-motions for summary judgment filed by the parties in this lawsuit, the court appeared to have little trouble determining that the actions involved here were ultra vires. It found that DOGE lacked any statutory authority over NEH programs but in practice assumed a central role in identifying grants for termination.

“The National Foundation on the Arts and the Humanities Act does not authorize DOGE to do what it did,” said the court. “The President’s Executive Orders do not authorize DOGE to do what it did. And even if the Executive Orders purported to do so, they could not supply the authority that Congress withheld. The President cannot confer upon an executive entity legal authority that neither Congress nor the Constitution has conferred. Where, as here, ‘there is no statute that expressly authorizes’ DOGE’s conduct, and no basis from which such authority ‘can fairly be implied,’ DOGE cannot act.”

First Amendment. As with the ultra vires analysis, the court likewise had little trouble finding First Amendment problems at play. “This case presents a textbook example of unconstitutional viewpoint discrimination,” it explained.

A key feature of DOGE’s review process—and a focal point of the court’s First Amendment analysis—was its reliance on artificial intelligence tools. Rather than conducting substantive evaluations of grant applications or completed work, DOGE personnel used AI-generated responses as justification for classifying grants. The court described this method as both cursory and arbitrary. It emphasized that the AI outputs were not independently verified, contextualized, or tested against the full content of the grants. In numerous instances, the classifications appeared tenuous or illogical, reflecting superficial keyword associations rather than meaningful analysis of the projects themselves. Nonetheless, these AI-assisted determinations became a significant basis for the sweeping termination decisions.

For example, the government, using ChatGPT, classified as DEI an anthology titled “In the Shadow of the Holocaust: Short Fiction by Jewish Writers from the Soviet Union.” The stated rationale was that the project “explores Jewish writers’ engagement with the Holocaust in the USSR, highlighting themes of witnessing, memory, and resilience.” It also classified as DEI a project examining graphic narratives created by Jewish women survivors of Nazi persecution in the immediate aftermath of the Holocaust because it “highlights the overlooked graphic narratives by Jewish women survivors, revealing diverse, gendered ways of experiencing and representing the Holocaust.”

No free pass for AI. Importantly, the court rejected any implication that the use of AI diminished governmental responsibility, noting that the government had suggested that there is no real constitutional problem here because any viewpoint-based classification was ChatGPT’s doing, rather than the responsibility of the government. However, the court explained that once DOGE officials adopted and relied upon AI-generated judgments, those judgments effectively became the government’s own. The court underscored that constitutional and statutory obligations cannot be avoided by delegating decision-making to automated tools, particularly where those tools are used in a mechanical or outcome-driven manner.

On constitutional grounds, the court concluded that the terminations violated the First Amendment. “Regardless of whether DOGE’s classifications were coherent (obviously, many of them were not),” the court explained, “what matters is that it deliberately sought to single out and eliminate grants based on its perception that they implicated disfavored ideas.”

Because NEH grants support the expression of private individuals and institutions rather than government speech, the selective cancellation of grants based on perceived viewpoints—particularly associations with DEI or a prior administration—constituted impermissible viewpoint discrimination. “In sum,” the court explained, “every aspect of the termination process—the NEH DEI ratings, DOGE’s DEI-based ChatGPT filter, and the spreadsheet sorting grants by ‘DEI Rationale’—reflects a single, unconstitutional purpose: to drive a particular disfavored viewpoint from the marketplace of humanistic inquiry.”

Equal protection. The court also identified a Fifth Amendment equal protection violation, noting that the classification process explicitly relied on categories such as race, gender, and ethnicity, triggering heightened scrutiny that the government could not satisfy. “In hundreds of instances in the record, the resulting DEI Rationale makes clear that ChatGPT made the classification solely because a grant involved a particular race, gender, or other protected trait.” The court rejected the federal government’s argument that rational basis, rather than strict scrutiny, applied to these circumstances, where a DOGE employee affirmatively endorsed ChatGPT’s race-based classifications and where the record reflected “systematic disparities” in the government’s treatment of projects involving different national origins and genders.

The court explained:

“Until now, it had not been the Court’s experience that the United States Attorney’s Office in this district would advance an argument that was so blatantly contradicted by the evidence—but it certainly has done so here. The record establishes, beyond any dispute, that the Government used protected characteristics as criteria for identifying grants for termination. DOGE swept in race and ethnicity—including grants concerning Black, Asian, Latino, and Indigenous communities—as well as national origin and immigration status; religion and religious identity (including Jewish, Christian, and Muslim subjects); sex; and sexual orientation, as criteria for grant termination. Rational basis review does not apply to the racial, ethnic, national-origin, or religious classifications; strict scrutiny does. Nor does rational basis review apply to sex-based classifications, which require an ‘exceedingly persuasive justification.’”

SCOTUS voting decision invoked. The court also pointed to the very recent U.S. Supreme Court’s decision in Louisiana v. Callais, in which Justice Alito wrote that “the Constitutional almost never permits the Federal Government or a State to discriminate on the basis of race,” and that “Such discrimination triggers strict scrutiny.” The court here pointedly observed, “If strict scrutiny applies with such force to race-conscious districting undertaken in purported compliance with the Voting Rights Act—that is, to an effort designed by Congress to preserve equal electoral opportunity and prevent racial exclusion from the political process—then it must apply with at least equal if not more force here, where race was used to exclude disfavored scholarship about historically marginalized groups from public support.”

In concluding, the court framed the episode as a cautionary example of how rapid, technology-assisted decision-making—especially when combined with political directives and institutional overreach—can undermine legal safeguards. The use of AI, rather than introducing neutrality or efficiency, in this instance amplified arbitrariness and facilitated constitutional violations, reinforcing the principle that emerging tools must remain subject to established legal constraints.

The court determined that the plaintiffs were entitled to equitable relief and that the case was properly brought in district court. It granted summary judgment in favor of the plaintiffs, certified a class of affected grantees, and issued a permanent injunction vacating the terminations.

The cases are Nos. 25-cv-3657 (CM) and 25-cv-3923 (CM).

News: LitigationNewsTrends WhiteHouseNews ESGNews AgencyNews IndividualRights RaceDiscrimination SexDiscrimination NationalOrigin Discrimination CoverageLiability DOGE FedTracker LaborEmployment AINews GCNNews