The SDNY’s NEH decision shows how anti-DEI enforcement can collapse when political labeling replaces statutory authority, individualized review, and constitutional limits.
Working Definitions
In this commentary, diversity, equity, and inclusion (“DEI”) refers to the broad set of policies, programs, and institutional practices commonly associated with expanding access, representation, and participation for historically excluded or underrepresented groups. In public debate, the term is often used imprecisely. That imprecision becomes dangerous when government officials use “DEI” as a screening label for official action.
Diversity, equity, inclusion, and accessibility (“DEIA”) is a related term that adds accessibility, usually referring to access for persons with disabilities and broader institutional participation. In the consolidated National Endowment for the Humanities actions, DEI and DEIA mattered because the government allegedly used those categories, along with “environmental justice” and “gender ideology,” to identify grants for termination.
The National Endowment for the Humanities (“NEH”) is the federal agency Congress created to support humanities scholarship, cultural preservation, public history, literature, language, philosophy, jurisprudence, ethics, archaeology, comparative religion, and related fields. NEH is not merely a funding office. It operates under a statutory structure Congress enacted, including a chairperson, the National Council on the Humanities, and a grant-review process involving expert and institutional review.
The United States Department of Government Efficiency Service (“DOGE”) refers to the executive-branch initiative involved in reviewing federal contracts and grants under the administration’s cost-efficiency directives. In these consolidated actions, the central issue was not merely that DOGE participated in review. The issue was that the court found DOGE personnel exercised decisive authority over which NEH grants would be terminated, despite lacking statutory authority to do so.
Ultra vires means action taken beyond lawful authority. When a government actor does something Congress did not authorize that actor to do, the action may be ultra vires. Here, the court held that DOGE’s role in selecting and directing NEH grant terminations was ultra vires because Congress vested NEH grant authority elsewhere.
Core Thesis
The consolidated NEH grant-termination actions are not merely another skirmish in the national fight over diversity, equity, and inclusion. They are a warning about what happens when government officials convert a political slogan into an enforcement mechanism. Once “DEI” became the filter, the process began treating race, gender, culture, identity, religion, historical memory, and disfavored scholarship as suspect categories. That is where the government’s position began to collapse.
The problem was not simply that the administration opposed DEI. Administrations change. Policy priorities change. Federal agencies may adjust programs within lawful limits. But government may not use ideological labels to punish protected viewpoints, target protected characteristics, displace statutory decisionmakers, or erase previously awarded grants through a process Congress did not authorize.
That is the lesson of the Southern District of New York’s ruling in American Council of Learned Societies, et al. v. National Endowment for the Humanities, et al., No. 25-cv-3657 (CM), consolidated with The Authors Guild, et al. v. National Endowment for the Humanities, et al., No. 25-cv-3923 (CM) (S.D.N.Y. May 7, 2026). These were two consolidated actions. The first was brought by major scholarly organizations. The second was brought by The Authors Guild and individual NEH grantees. Together, they challenged the same government program: the mass termination of previously awarded NEH grants through a process allegedly driven by DOGE, political classifications, keyword searches, abbreviated spreadsheets, and ChatGPT-generated “DEI rationales.”
The court did not rescue DEI as a political brand. It rejected government by label. It rejected the use of a vague ideological category as a substitute for statutory authority, individualized review, constitutional discipline, and lawful decision-making. That distinction matters. Civil-rights law does not exist to protect one ideology against another. It exists to restrain government power when government targets people, speech, scholarship, identity, association, or viewpoint through unlawful means. In these consolidated actions, the anti-DEI campaign did not merely dismantle programs. It allegedly created the evidentiary record that dismantled the government’s own defense.
I. The Court Did Not Rescue DEI. It Rejected Government by Label.
The easiest way to misunderstand the consolidated NEH grant-termination actions is to reduce them to a culture-war headline. That would be analytically lazy and legally wrong. The ruling is not a judicial declaration that every program bearing a DEI label is lawful, wise, necessary, or immune from reform. It is not a ruling that a new administration must preserve the policy choices of the administration that came before it. Nor is it a ruling that federal grant recipients have a constitutional entitlement to public money forever.
The ruling is more important than that.
The court rejected the government’s attempt to convert a political label into a legal weapon. It rejected a termination process that treated the phrase “DEI” as if it were self-defining, self-executing, and legally sufficient. It rejected the idea that an administration can take a contested political term, attach it to disfavored scholarship, and use that label to cancel previously awarded grants without the process, authority, and constitutional restraint Congress and the Constitution require.
That is the real significance of these consolidated actions. They expose the legal danger of governing through slogans.
In politics, DEI can mean almost anything depending on who is speaking. To some, it means unlawful preferencing. To others, it means remedial inclusion. To others, it means bureaucratic excess. To others, it means the basic study of race, gender, culture, disability, religion, immigration, sexuality, inequality, exclusion, or institutional history. Political actors often benefit from that ambiguity. Courts do not.
In litigation, ambiguity becomes evidence. If the government uses a term as a basis for official action, the term must do legal work. It must be defined. It must be tethered to statutory authority. It must be applied through a lawful process. It must distinguish lawful targets from unlawful ones. It must not operate as a proxy for protected identity, protected viewpoint, protected association, or disfavored speech.
That is where the government’s anti-DEI theory began to work against itself.
The two consolidated actions challenged what the court described as the largest mass termination of previously awarded grants in NEH history. More than 1,400 grants were terminated, involving over $100 million in congressionally appropriated funds awarded to scholars, writers, research institutions, and humanities organizations. The American Council of Learned Societies (“ACLS”) action was brought by ACLS, the American Historical Association (“AHA”), and the Modern Language Association of America (“MLA”). The Authors Guild action was brought by The Authors Guild and individual NEH grantees whose awards were swept into the same termination process.
That procedural posture matters. These were not abstract objections filed by spectators in the national DEI debate. These were consolidated actions brought by institutions and individuals alleging direct injury from a concrete government program. The institutional plaintiffs alleged harm to themselves and their members. The individual grantees alleged disruption of funded scholarly work, loss of expected grant support, and professional consequences flowing from termination decisions made after they had already structured work, leave, research, and income around awarded grants.
The government defended the terminations as part of a presidential effort to eliminate grants associated with DEI, DEIA, environmental justice, and gender ideology, while reducing discretionary spending and advancing the new administration’s priorities. But that defense carried a built-in litigation risk. Once the government acknowledged that it was sorting grants through those categories, the decisive question became: what did those categories actually mean in operation?
Were they being used to identify unlawful racial quotas? Were they being used to detect grantee noncompliance? Were they being used to enforce specific grant terms? Were they being used to remedy fraud, waste, or fiscal abuse? Or were they being used to target scholarship because it involved race, gender, sexuality, religion, Indigenous history, Black newspapers, immigrant communities, women’s intellectual life, public health activism, or other identity-linked subjects?
That distinction was fatal. A government can change priorities. It can reduce spending within lawful bounds. It can decline to fund future projects that no longer fit statutory and policy objectives. It can enforce grant terms when a recipient fails to comply. But it cannot use public funding power to punish ideas, purge disfavored viewpoints, or target subjects because they are associated with protected identity.
The court’s record shows why the DEI label became so unstable. DOGE personnel used keyword searches and categories such as “Craziest Grants” and “Other Bad Grants.” The detection terms included words and concepts such as “Tribal,” “Immigrants,” “diversity,” “inclusion,” “Indigenous,” “Native,” “Equity,” “Equality,” “Marginalized,” “BIPOC,” “Solidarity,” “Citizenship,” “Melting Pot,” “Social Justice,” and “Gay.” Those are not neutral indicators of illegality. Many are ordinary descriptors of American history, culture, identity, social conflict, and constitutional life.
The examples matter because they show the difference between political suspicion and legal analysis.
A project involving Native American communities could be treated as DEI because it involved Indigenous knowledge, culture, climate, and stereotypes. A project digitizing African-American newspapers could be characterized as DEI because it preserved underrepresented voices. A project about American women in Paris could be swept into the same frame because it concerned women’s cultural and intellectual life. A project on Uyghur persecution could be treated as DEI because it involved an ethnic and religious minority facing state repression. A project involving ancient Jewish texts could be pulled into the classification system because it involved Jewish thought. A project concerning plastics, municipal regulation, and urban infrastructure could somehow become DEI because of generalized language about government influence and social impact.
The problem was not merely that the classifications were broad. The problem was that they collapsed subject matter, identity, history, and viewpoint into a single politically disfavored category.
That is not legal review. That is ideological sorting.
This is why the ruling should not be read as a judicial endorsement of DEI. It should be read as a judicial rejection of category-driven punishment. The court’s point was not that every grant concerning race, gender, culture, or identity must survive forever. The point was that the government may not make those subjects suspect by label and then impose mass consequences without lawful authority, individualized review, and constitutional justification.
That is the civil-rights significance of the ruling. Civil-rights law is not a one-way ratchet. It restrains government when government discriminates in the name of inclusion. It also restrains government when government retaliates in the name of dismantling inclusion. The constitutional injury does not depend on whether the slogan is progressive or conservative. It depends on whether the state uses power to punish protected expression, protected identity, protected association, or disfavored viewpoint.
The court concluded that plaintiffs were entitled to judgment as a matter of law on claims that the mass terminations violated the First Amendment, violated the equal-protection component of the Fifth Amendment, and were ultra vires because DOGE officials exercised decisive authority over NEH grants without statutory authority. That combination is important. The First Amendment addressed viewpoint and expressive injury. Equal protection addressed protected-characteristic targeting. Ultra vires doctrine addressed lawful power. Together, those doctrines exposed the same structural defect from different directions: the government used the wrong decisionmaker, the wrong process, and the wrong criteria.
That is why the consolidated actions belong in a broader public conversation about government power. The court did not say government must fund every idea. It said government cannot hide punishment behind a label. It said executive officials cannot take a contested political phrase, refuse to define it with legal discipline, apply it through mechanical screening, and then use it to erase grants Congress authorized and agencies awarded.
The lesson is not that DEI survived. The lesson is that constitutional government survived the attempted conversion of anti-DEI politics into administrative machinery.
II. The Statutory Scheme Made the Shortcut Unlawful
The government’s constitutional problem was inseparable from its statutory problem. Congress did not create NEH as an open account to be reprogrammed by political operatives. It created a statutory institution. It assigned authority. It identified permissible purposes. It created procedural safeguards. It placed grantmaking within a structure of expertise, recommendation, and accountable agency action.
That structure was not decorative. It was the law.
The consolidated actions therefore cannot be understood solely as a dispute over DEI. They were also a dispute over whether executive actors may override a congressionally designed grant program because a new administration disapproves of prior funding choices. The answer was no.
The court’s statutory analysis begins with a basic separation-of-powers premise: executive action cannot create authority Congress did not grant. An executive order may direct the exercise of existing authority, organize executive-branch operations, or set lawful priorities. It cannot rewrite a statute. It cannot transfer statutory power from one decisionmaker to another. It cannot authorize an advisory entity to do what Congress assigned to a Senate-confirmed agency head or to a statutory process.
That principle controlled the NEH dispute.
Congress established NEH within the National Foundation on the Arts and the Humanities. Congress provided that NEH would be headed by a chairperson appointed by the president with Senate confirmation. But presidential appointment did not convert NEH’s statutory funds into personal executive property. Congress, not the president, created the grant program. Congress, not DOGE, appropriated the funds. Congress, not a political advisory team, identified who could exercise NEH funding authority and under what conditions.
The statutory structure was specific. The NEH chairperson was authorized to enter into grants and other forms of assistance, but that authority had to be exercised “with the advice of the National Council on the Humanities.” The Council had statutory duties to review applications and make recommendations. The chairperson could not approve or disapprove covered applications until receiving the Council’s recommendation, subject to limited exceptions.
That means Congress did not design NEH grantmaking as raw political discretion. It designed a process. Peer review, staff assessment, Council involvement, and final agency action were not bureaucratic ornaments. They were the mechanism Congress selected to protect public funding decisions in a field where intellectual freedom, cultural pluralism, scholarly judgment, and viewpoint neutrality matter.
The government’s error was treating that statutory architecture as expendable once a new president announced different priorities.
The statutory purposes made that error even more glaring. Congress did not merely authorize NEH to fund generic scholarship. It expressly recognized that the humanities serve public purposes tied to cultural heritage, intellectual inquiry, diverse beliefs and values, and access to knowledge. Congress authorized support for programs and research that have scholarly and cultural significance and that reflect the diversity and richness of American cultural heritage, including minority, inner-city, rural, and tribal communities. The statute itself placed diversity of belief, culture, community, history, and access inside NEH’s mission.
That does not mean the phrase DEI is automatically protected. It does not mean every modern DEI program fits NEH’s statute. It does not mean an agency must fund any project merely because it invokes identity, access, or representation. But it does mean that the government could not lawfully treat the mere presence of those subjects as evidence of statutory disqualification. Congress had already made many of those subjects part of the statutory domain.
This is where the government’s anti-DEI enforcement theory became legally self-defeating. If the statutory mission includes scholarship about minority communities, tribal communities, diverse beliefs, cultural heritage, and public understanding, then the government cannot use broad ideological hostility to those subjects as a substitute for statutory analysis. The question is not whether a grant touches race, gender, religion, national origin, culture, or identity. The question is whether the grant falls within the statutory purposes, complies with grant terms, and was administered through the decision-making structure Congress required.
That is a very different inquiry from asking whether something “relates at all to DEI.”
The post-award provisions made the shortcut even less defensible. The court emphasized that the statute did not authorize wholesale revocation of already awarded grants because a new administration disagreed with earlier grant decisions or preferred to redirect funds elsewhere. Post-award consequences were tied to grantee-specific compliance findings, financial reporting, project reporting, and whether a recipient substantially failed to satisfy the purposes for which assistance was provided.
That distinction matters. The government was not merely deciding what to fund next year. It was terminating grants already awarded. Those awards had grant periods. They had recipients. They had project expectations. They had reliance consequences. Some grantees had taken leave, declined other opportunities, structured scholarly work, or depended on scheduled payments. In the Authors Guild action, the injury was not institutional in the abstract; it was professional, financial, and practical.
Previously awarded grants cannot be treated as if they are nothing more than political preferences waiting to be reversed. Once the government awards a grant under a statutory program, post-award action must be tied to the legal framework governing that award. If a grantee misuses funds, fails to comply with conditions, abandons the project, or violates material terms, the government has tools. But those tools are individualized. They are record-based. They are tied to compliance. They are not a license for ideological cancellation.
The court’s contrast between NEH’s historical practice and DOGE’s process is important. NEH historically did not terminate grants in this manner. The record reflected that terminations were rare and typically involved grantee noncompliance. There was no historical practice of canceling the vast majority of already awarded grants because a new administration disagreed with the subjects being studied or written about. That contrast gave the court a practical measure of statutory deviation.
The ordinary NEH model was expert-driven, layered, and individualized. Grant proposals moved through peer review, staff recommendations, Council review, and final chairperson action. The DOGE model was the opposite. It relied on spreadsheet descriptions, keyword searches, subjective political classifications, and AI-generated rationales. The statutory model asked whether grants satisfied congressional purposes and agency standards. The DOGE model asked whether projects related “at all” to DEI. The statutory model centered the NEH chairperson and National Council. The DOGE model displaced them. The statutory model required lawful authority. The DOGE model assumed it.
That is why the shortcut was unlawful.
DOGE’s process rested on abbreviated descriptions and a binary ChatGPT prompt asking whether each project “relate[d] at all to DEI.” The resulting AI-generated rationales were folded into final spreadsheets used to identify grants for termination. The court observed that nothing in the record suggested NEH had ever used such a process before March 2025, and nothing in the authorizing statute permitted it.
That sentence captures the administrative-law failure. Novelty is not automatically illegality. Agencies may modernize. They may use technology. They may change internal workflows. But they may not replace statutory criteria with nonstatutory political categories. They may not replace expert review with ideological word searches. They may not replace individualized compliance determinations with categorical suspicion. They may not replace a congressional decisionmaker with an advisory team lacking statutory authority.
This is where the consolidated actions become a separation-of-powers warning. Congress vested funding authority in NEH’s statutory structure. DOGE substituted presidential policy judgment for that statutory structure. The court treated that as executive lawmaking, not faithful execution.
That should be the central lesson of Section II: process is not paperwork when government power is being used to punish disfavored ideas. Statutory structure is not red tape. It is the legal boundary between democratic governance and executive improvisation. The more politically charged the subject, the more important the process becomes.
If a new administration believes a prior administration funded unlawful grants, it may act. But it must act lawfully. It must identify the statutory violation, the grant-term violation, the compliance defect, or the lawful prospective policy change. It must use the decisionmaker Congress authorized. It must make individualized determinations where the statute requires individualized determinations. It must respect constitutional limits.
What it cannot do is treat DEI as a magic word that dissolves statutory law.
III. DOGE Was Not Treated as an Adviser. It Was Treated as the Decisionmaker.
The DOGE issue gives the consolidated actions their institutional force. Without it, the dispute could be mistaken for an ordinary disagreement over federal funding priorities. With it, the dispute becomes something larger: an unauthorized transfer of operational power from a statutory agency to a political advisory apparatus.
That is why the court’s treatment of DOGE matters. The court did not merely say DOGE was involved. It did not merely say DOGE influenced the process. It found, on the record before it, that DOGE exercised decisive authority over which NEH grants would be terminated. That distinction is everything.
The government needed DOGE to be advisory. That was the only way to fit DOGE’s role inside the statutory structure. Executive Order 14222 directed agency heads, in consultation with agency DOGE team leads, to review existing covered contracts and grants and, where appropriate and consistent with law, terminate or modify them. The operative words were “agency heads.” DOGE’s role, at least on paper, was consultation.
The record told a different story.
The court emphasized that operative authority was supposed to remain with agency heads and that DOGE’s role was expressly advisory. It then concluded that the record demonstrated DOGE’s role was anything but advisory. The court stated plainly that DOGE made the decisions for NEH.
That finding should be central to the commentary because it strips away the formalism that often protects unlawful government action. Government defendants frequently rely on paperwork: whose name appeared on the notice, whose title appears in the regulation, whose signature block appeared at the bottom of the letter. But courts are not required to worship form when the record shows a different operational reality. If the statutory decisionmaker merely lent his name to a decision made elsewhere, the signature does not cure the defect. It proves the displacement.
Here, the court’s description of DOGE’s role was stark. Before ever meeting with NEH leadership, Fox had already started reviewing NEH grants using keyword searches and DOGE-generated categories to identify grants he considered objectionable. Those categories included “Craziest Grants” and “Other Bad Grants.” DOGE then used ChatGPT-generated DEI rationales, combined them with NEH staff ratings, and generated the final spreadsheets used to identify grants for termination.
That is not consultation. That is pre-decisional targeting.
The problem was compounded by the background of the actors and the nature of the review. The record described DOGE personnel operating as part of a “Small Agencies Team.” The court noted that they lacked experience in government, grant administration, and humanities review. Yet they were reviewing grants awarded through NEH’s statutory process, using abbreviated descriptions rather than underlying applications or full grant records. They were not applying NEH’s expert-review model. They were not conducting grantee-specific compliance analysis. They were not asking whether recipients failed to satisfy grant purposes. They were sorting federally awarded humanities grants through politically generated categories.
The labels alone reveal the bias of the machinery. “Craziest Grants” is not a legal category. “Other Bad Grants” is not a statutory standard. “Detection Codes” is not congressional authorization. Those terms may be useful in a political campaign or an internal ideological screening exercise. They have no place as decisional criteria for terminating previously awarded federal grants unless grounded in lawful standards. The court’s record showed no such grounding.
The ChatGPT component made the defect more visible, but it did not create the defect by itself. Artificial intelligence was not the original sin. The original sin was unlawful delegation, nonstatutory criteria, and viewpoint-laden targeting. ChatGPT merely amplified the weakness of the process. It generated short “DEI rationales” in response to a prompt asking whether each project related “at all” to DEI. That prompt was overbroad by design. It did not define DEI. It did not ask whether the grant violated the law. It did not ask whether the grantee breached grant conditions. It did not ask whether the project fell outside NEH’s statutory mission. It asked whether the project related “at all” to a politically charged concept.
A prompt that broad will capture almost anything in the humanities. History relates to identity. Literature relates to culture. Religion relates to belief. Jurisprudence relates to citizenship. Philosophy relates to equality. Public health relates to community. Archives relate to whose voices are preserved. The humanities are, by statutory design, concerned with human experience. If “relates at all to DEI” becomes the test, then the test is not a test. It is a dragnet.
That is why the court’s examples matter. Projects involving Jewish texts, Uyghur persecution, plastics and municipal regulation, American women in Paris, African-American newspapers, Indigenous knowledge, and Holocaust education were swept into or considered within the termination machinery. Some of those subjects fall squarely within the ordinary work of humanities scholarship. Some reflect the very statutory purposes Congress assigned to NEH. Their presence in the termination process showed how far the anti-DEI label had drifted from any lawful administrative criterion.
The decisive evidence, however, was not only the classification process. It was DOGE’s control over the final result.
The court described evidence that DOGE personnel rejected the NEH acting chair’s recommendation to continue funding certain grants and added more grants for termination, including grants NEH staff had identified as “N/A,” meaning not applicable to the DEI-related categories under review. The court further stated that the appearance of the acting chair’s name on the termination letters did not alter the substance of what occurred because DOGE personnel drafted and sent the notices under the acting chair’s name.
That is the heart of the ultra vires problem. The statutory decisionmaker was displaced. The agency’s own recommendations were rejected by DOGE personnel. The final communications bore the appearance of NEH action, but the operational record pointed elsewhere.
The internal communications made the point impossible to ignore. Just days before the mass termination, DOGE allegedly pushed for a rapid “game plan” for grant terminations and contract cancellations, invoked pressure, and pressed NEH officials to remain “on our side.” The court concluded that this was not the language of an adviser awaiting a decision from the statutory decisionmaker.
That language is devastating because it reveals the institutional posture. DOGE was not asking NEH what NEH had decided. DOGE was telling NEH what DOGE intended to carry out. The statutory agency head was treated less like a decisionmaker and more like a participant whose cooperation was preferred but not necessary.
That is how advisory power becomes shadow power.
This is where the consolidated actions become larger than NEH, larger than DOGE, and larger than DEI. If an advisory team can enter an agency, classify programs through ideological screens, reject the agency’s own recommendations, direct termination lists, and send notices under the agency head’s name, then statutory design becomes a formality. Congress may vest authority in one place, but operational power will move to another. The public will see the agency’s signature, while the real decision is made elsewhere.
That is precisely what ultra vires doctrine exists to prevent. Government actors may not exercise binding authority over federal programs unless Congress gave them that authority. An advisory body may advise. It may coordinate. It may make recommendations. It may help an agency identify issues for lawful review. But it cannot become the decisionmaker. It cannot exercise grant-termination authority Congress vested in NEH’s statutory structure. It cannot use executive urgency to bypass statutory limits.
The court’s conclusion was direct: DOGE’s process displaced the statutory official responsible for NEH post-award administration, applied nonstatutory criteria, and selected grants for termination without the recipient-specific compliance findings contemplated by statute. That was not advice. It was the exercise of termination authority over NEH’s grant program, authority Congress vested elsewhere and never gave to DOGE.
That is the institutional lesson. The court did not simply reject a bad DEI-screening process. It rejected an unlawful power transfer. The government tried to defend the terminations as executive implementation of policy priorities. The record showed something more troubling: a statutory agency’s authority was functionally overtaken by actors without statutory power, using criteria Congress never enacted, through a process the statute did not permit.
That is why the consolidated NEH grant-termination actions should not be written as a narrow DEI commentary. They should be written as a warning about constitutional government under ideological pressure. Political movements always want speed. Statutory law often requires sequence. Political actors want blunt categories. Administrative law requires reasoned decision-making. Political messaging rewards slogans. Civil-rights law asks whether those slogans are being used to target protected speech, protected identity, or disfavored viewpoints.
Here, the government’s anti-DEI machinery did not merely fail to persuade the court. It allegedly exposed the very constitutional and statutory defects the plaintiffs needed to prove.
IV. Technology Did Not Cause the Abuse. It Scaled It.
One additional point should not be missed. The court’s discussion of ChatGPT is not merely a side detail or a colorful fact in the record. It is a warning about how modern technology can be misused when the underlying human objective is unlawful.
ChatGPT did not decide to terminate NEH grants. People did. ChatGPT did not create the anti-DEI enforcement campaign. People did. ChatGPT did not define the legal standard, select the targets, reject agency recommendations, or send termination notices. People did those things. The danger exposed by the consolidated NEH actions is more precise: government actors allegedly used artificial intelligence to accelerate, rationalize, and cosmetically legitimize a process that was already infected by political targeting, statutory disregard, and constitutional risk.
That distinction matters. Technology can assist lawful review. It can organize information. It can identify patterns. It can help agencies manage large records. But when officials begin with an improper objective, technology can magnify the abuse. A biased human process does not become neutral because it is routed through software. A vague political label does not become a legal standard because an artificial-intelligence tool generates a sentence around it. An unlawful termination does not become reasoned agency action because a spreadsheet contains a short “rationale.”
The record described a process in which abbreviated grant descriptions were fed into ChatGPT with a prompt asking whether each project related “at all” to DEI. That prompt did not define DEI. It did not ask whether the grant violated the law. It did not ask whether the recipient breached grant terms. It did not ask whether the project fell outside the statutory purposes of NEH. It asked a broad, politically charged question and then produced short rationales that were folded into the termination machinery.
That is how technology can be misused to discriminate. Not because the technology has independent legal intent, but because human actors can use it to create the appearance of analysis while avoiding actual judgment. They can make discriminatory targeting look administrative. They can make viewpoint suppression look data-driven. They can make ideological sorting look efficient. They can use speed as a substitute for legality.
That is the caution for lawyers, courts, agencies, and the public. Artificial intelligence is not inherently unlawful. But it is not inherently neutral either. Its legal significance depends on who is using it, what they are asking it to do, what criteria they supply, what records they ignore, and whether the output is being used to support government action affecting rights, funding, employment, liberty, or public benefits.
In the consolidated NEH actions, the misuse was not subtle. ChatGPT was allegedly used to generate “DEI rationales” for grants that DOGE personnel had not reviewed through the statutory process Congress required. That made the technology part of the evidentiary trail. It showed how a political objective could be converted into an administrative record through automation.
That is why courts must remain especially alert when government defendants invoke technology as part of the decision-making process. The question is not whether a tool was used. The question is whether the tool was used lawfully. What was the prompt? What was the data source? What criteria were supplied? Who reviewed the output? Was the output tested against the governing statute? Did a lawful decisionmaker exercise independent judgment? Were affected parties given a meaningful explanation? Or was technology used to create post hoc justification for a predetermined result?
V. The First Amendment Problem: Viewpoint Discrimination by Grant Termination
The First Amendment problem in the consolidated NEH grant-termination actions was not that the government refused to fund every scholarly idea. That would be too broad and too easy to dismiss. The government is not required to subsidize every project, every writer, every scholar, every institution, or every field of study. Federal grant programs have limits. Agencies make choices. Congress defines priorities. Administrations may shift emphasis within lawful bounds.
The constitutional problem begins somewhere else: once the government creates a statutory grant program, awards grants through that program, and then terminates already-awarded grants because of the perceived viewpoint, subject matter, identity association, or expressive content of the work, the dispute no longer remains a routine funding matter. It becomes a speech problem.
That is what made the government’s position vulnerable. The terminations did not arise from ordinary grant administration. They did not arise from individualized findings that recipients misused funds, abandoned projects, violated material grant conditions, or failed to perform. The record described a mass process that used political classifications, keyword searches, ChatGPT-generated “DEI rationales,” and DOGE-driven termination lists to cancel previously awarded grants. The court concluded that the mass termination violated the First Amendment and that the injuries included disruption of protected expression, interruption of ongoing research and publication, cancellation or suspension of humanities programming, and a chilling effect caused by viewpoint-based and unauthorized criteria.
That is the correct frame. A grant termination can be more than a financial event. It can be an expressive injury.
The humanities are not the same as a road-construction contract or a paper-supply agreement. NEH grants fund scholarship, research, writing, archives, education, publication, preservation, and public programming. Those activities are communicative. They involve ideas, interpretation, memory, argument, culture, identity, historical narrative, and public understanding. When the government terminates those grants because it dislikes the perceived viewpoint of the project, it is not merely reallocating dollars. It is interfering with the production and dissemination of ideas.
That is why the “DEI” label became so dangerous in this record. If the government had identified specific grants that violated statutory criteria, grant terms, or lawful restrictions on discrimination, the analysis would look different. But a broad assertion that a grant relates “at all” to DEI does not establish illegality. It does not establish misuse of federal funds. It does not establish nonperformance. It does not establish that the project falls outside NEH’s statutory mission. In the humanities context, a project may relate to race, gender, religion, national origin, sexuality, disability, Indigenous history, immigration, equality, inequality, or cultural access because those subjects are part of human experience and public life.
The government’s theory therefore risked converting subject matter into suspicion. A project about Black newspapers could be treated as suspect because it preserved underrepresented voices. A project involving Native American communities could be treated as suspect because it addressed Indigenous culture. A project concerning women’s intellectual life could be treated as suspect because it centered women. A project involving Uyghur persecution could be treated as suspect because it concerned an ethnic and religious minority. In that posture, “DEI” did not function as a precise legal category. It functioned as a viewpoint screen.
That is the difference between lawful funding discretion and unconstitutional viewpoint discrimination. The government may define the boundaries of a grant program. It may fund some topics and not others prospectively, if it acts within statutory and constitutional limits. But it may not punish a funded project because the government disfavors the project’s perspective on race, gender, history, identity, religion, sexuality, culture, or public memory.
The court’s permanent-injunction discussion made the expressive harm explicit. It explained that monetary damages would not adequately remedy the harm because the injury was not limited to the loss of money. The injury included the disruption of protected expression and the chilling effect caused by the government’s use of viewpoint-based and unauthorized criteria to terminate federal grants. The court also stated that the government had no cognizable interest in enforcing terminations that were unauthorized by statute and infected by viewpoint discrimination.
That language matters. The court was not treating the plaintiffs as disappointed vendors. It was treating them as speakers, scholars, writers, institutions, and associations whose expressive work had been disrupted by government action.
This is where the consolidated posture of the two actions becomes important. The ACLS plaintiffs brought the institutional and associational dimension. They represented scholarly organizations whose members depended on NEH funding for humanities work. The Authors Guild action brought the individual dimension. It included writers and scholars whose specific projects were terminated after grants had already been awarded. Together, the two actions showed the full First Amendment injury: not only the disruption of individual projects, but also the chilling of a broader ecosystem of scholarship and public expression.
The chilling effect is not abstract. If scholars, writers, archives, universities, museums, and cultural institutions learn that projects involving race, gender, Indigenous history, religion, sexuality, immigration, or minority communities may be placed on a political termination list, they will adjust behavior. They will soften titles. They will avoid words. They will recode descriptions. They will steer away from controversial histories. They will preemptively sanitize scholarship to avoid becoming the next target.
That is precisely why viewpoint discrimination is so corrosive. It does not merely punish the speaker already targeted. It sends a warning to everyone else.
The government’s use of mass termination intensified the constitutional problem. An individualized grant-compliance review may at least show that the agency looked at the project, reviewed the award terms, considered the recipient’s conduct, and tied the action to lawful criteria. A mass termination based on broad ideological screens sends a different message. It tells the field that certain ideas are disfavored, certain histories are dangerous, and certain identities may trigger government sanction when they appear in funded work.
That is not neutral administration. That is expressive control.
The First Amendment analysis also exposes why the government’s “funding discretion” defense could not carry the day. Funding discretion is not a constitutional blank check. The government may choose what it funds, but it cannot use funding as a lever to impose unconstitutional conditions, penalize protected expression, or purge disfavored viewpoints from a statutory program. Once the government uses a grant termination to punish expression, the fact that money is involved does not eliminate the constitutional injury.
The court’s ultimate relief reflected that understanding. It declared the mass termination unlawful, unconstitutional, ultra vires, and without legal effect. It permanently enjoined the government from enforcing or giving effect to the mass termination and from terminating those grants on the same or substantially similar impermissible bases.
That remedy is important because it did not prevent NEH from administering its grants lawfully. It did not freeze all future agency discretion. It did not require the government to fund unlawful activity. It simply barred enforcement of the mass termination and barred repetition of the same impermissible grounds. In other words, the court did not remove government authority. It restored legal boundaries around that authority.
That is the First Amendment lesson: the government cannot launder viewpoint discrimination through budget language, efficiency language, or anti-DEI language. If the real target is protected expression, the label does not save the action. It helps prove the violation.
VI. The Equal Protection Problem: When Anti-DEI Becomes Identity Targeting
The equal-protection problem is narrower than the public debate and more serious than the government’s rhetoric allowed. The issue is not whether every anti-DEI policy is unconstitutional. It is not whether the government may review programs for unlawful preferencing, unlawful quotas, or impermissible discrimination. It can. Government may identify and end unlawful discrimination. It may insist that federal funds comply with statutory and constitutional limits.
The issue is what happens when anti-DEI enforcement becomes a mechanism for targeting race, gender, sexuality, religion, national origin, Indigenous status, immigrant identity, or other protected or identity-linked subjects. At that point, the government’s stated opposition to discrimination can become evidence of discrimination.
That is the pivot in these consolidated actions.
The court described plaintiffs’ theory as one involving discrimination on the basis of viewpoint and protected characteristics by targeting projects associated with “diversity, equity, and inclusion,” “gender ideology,” and other subject matter the administration viewed as inconsistent with its priorities. The Authors Guild plaintiffs specifically alleged that grants were identified for termination using criteria tied to race, gender, and subject matter, including keyword searches and automated artificial-intelligence tools.
That matters because equal protection is concerned with government classification and motive. The government does not avoid equal-protection scrutiny merely by changing labels. If the practical effect of a screening system is to target projects because they concern Black communities, Indigenous communities, women, LGBTQ subjects, immigrants, religious minorities, or other protected or identity-linked groups, the government cannot simply say it was targeting “DEI” and end the inquiry.
Labels do not control legal reality. Function does.
That is why the keyword evidence is important. Terms such as “BIPOC,” “Indigenous,” “Native,” “Tribal,” “Gay,” “Immigrants,” “Equity,” “Equality,” “Marginalized,” and “Social Justice” are not inherently unlawful. They do not prove fraud. They do not prove grant misuse. They do not prove nonperformance. They do not prove violation of federal law. In many humanities projects, those terms may identify the subject of legitimate scholarship. They may describe the people being studied, the archives being preserved, the communities being documented, or the historical wrong being examined.
When government treats those terms as detection codes for termination, the equal-protection risk becomes obvious. The state is not merely rejecting a theory of institutional management. It may be targeting identity itself.
The danger is especially acute in the humanities because identity and subject matter often overlap. A project on African-American newspapers is about race, history, journalism, archives, public memory, and democratic participation. A project on Native communities is about Indigenous history, culture, sovereignty, geography, ecology, and public understanding. A project on women in Paris is about gender, literature, art, migration, intellectual life, and transnational history. A project on Uyghur persecution is about religion, ethnicity, state repression, human rights, and cultural survival.
If the government treats those subjects as presumptively suspect because they resemble DEI, then anti-DEI enforcement begins to function as identity suppression.
That is not a minor evidentiary problem. It is the central civil-rights problem.
The Fifth Amendment’s equal-protection component applies to the federal government. It bars the federal government from discriminating on protected grounds. The question is not only whether the government used explicit racial or gender classifications in a formal regulation. Courts also look at substance: what criteria were used, what groups were targeted, what evidence reveals motive, whether the stated justification matches the record, and whether supposedly neutral terms operate as proxies for protected characteristics.
Here, the record described a process in which DOGE personnel used broad ideological categories, keyword searches, and ChatGPT-generated rationales to identify grants for termination. The court ultimately concluded that the mass termination violated the equal-protection component of the Fifth Amendment.
That conclusion should be understood as a warning. Anti-DEI enforcement can be lawful when it targets unlawful conduct. It becomes legally dangerous when it targets people, communities, identities, or scholarship because those subjects are politically disfavored. The government cannot take the language of equality law and invert it into a tool for suppressing equality-related speech, minority history, gender-based scholarship, religious minority study, or cultural preservation.
That is the backfire.
The administration’s stated objective was to dismantle DEI. But the process allegedly used to do so created evidence that the government was targeting protected characteristics and identity-linked subjects. The more the government relied on terms associated with race, gender, sexuality, religion, Indigenous status, and marginalized communities, the more it invited the question: was the government attacking unlawful discrimination, or was it attacking the study and visibility of protected groups?
That question is not rhetorical. It is legal.
The record also shows why artificial intelligence made the equal-protection issue more troubling. ChatGPT was asked whether projects related “at all” to DEI. That prompt did not distinguish between unlawful discrimination and lawful scholarship about historically excluded communities. It did not distinguish between a discriminatory admissions quota and a historical study of Black newspapers. It did not distinguish between compelled ideology and a project about women writers. It did not distinguish between unlawful preferencing and cultural preservation. A prompt that blunt cannot produce legally disciplined equal-protection analysis. It can only amplify the bias built into the human question.
That is how technology can assist identity targeting without ever announcing discriminatory intent. Human actors select the terms. Human actors define the target. Human actors decide what outputs matter. Human actors fold those outputs into government action. The software becomes the accelerant, not the origin.
For a civil-rights audience, this point is essential. Discrimination in modern government action will not always appear as explicit racial exclusion or openly discriminatory language. It may appear as proxy criteria. It may appear as algorithmic screening. It may appear as a spreadsheet category. It may appear as a facially neutral “efficiency” review. It may appear as a political campaign against “waste.” The task is to ask what the criteria actually do.
In these consolidated actions, the criteria allegedly did more than eliminate unlawful programs. They swept broadly across scholarship connected to identity, culture, history, and protected expression. That is why the equal-protection claim had force.
The government’s likely public defense would sound simple: we were ending discriminatory DEI. But the legal response is equally simple: prove it. Identify the unlawful practice. Identify the grant term violated. Identify the statutory prohibition. Identify the recipient-specific conduct. Identify the lawful decisionmaker. Identify the record showing that the termination was based on illegality rather than identity-linked viewpoint or subject matter.
That is what the process failed to do.
Equal protection does not prevent government from enforcing equality law. It prevents government from using equality rhetoric as cover for unequal treatment. That is the distinction the commentary must preserve. The point is not that DEI is untouchable. The point is that people, communities, subjects, and viewpoints do not become fair game because a political actor labels them DEI.
The consolidated actions therefore mark an important civil-rights inversion. The government attempted to use anti-DEI language as a shield. The record turned that language into evidence. When anti-DEI enforcement targets protected identity or the study of protected identity, it does not dismantle discrimination. It risks becoming discrimination.
VII. The Administrative Law Problem: Arbitrary, Capricious, and Pretextual Decision-Making
The administrative-law problem is the most practical defect in the consolidated NEH actions because it shows how the government’s process failed before the court even reached the larger constitutional stakes. Administrative law requires government to give real reasons for real decisions. It requires a decisionmaker to consider relevant facts, apply lawful criteria, explain the action taken, and produce a record that allows judicial review. It does not permit agencies to offer a clean public explanation while concealing the actual machinery that drove the decision.
That is where the mass termination process broke down.
The court’s discussion of the administrative record is unusually important. The government initially produced a record that did not contain the materials needed to evaluate the real decision-making process. The court found that the original record contained no contemporaneous explanation of how grants were selected for termination, who applied the selection criteria, or what information was reviewed. It also lacked materials reflecting keyword screening, automated processes, generative artificial-intelligence tools, and the involvement of actors outside NEH.
That is not a paperwork problem. That is a judicial-review problem.
A court cannot evaluate agency action if the government withholds the materials that actually informed the action. Judicial review becomes theater if the administrative record excludes the criteria, decisionmakers, methods, and communications that produced the result. That is why supplementation became necessary. After further production, the government produced thousands of additional documents, including materials showing the criteria and methods used to identify grants for termination, keyword screening, ChatGPT use, and the role of actors outside NEH. The supplemental production confirmed that the original record was incomplete and that it omitted the materials that actually informed the mass termination.
That point should not be lost. The administrative record itself became a site of institutional failure.
The government’s stated justifications could not be tested without knowing what actually happened. Who selected the grants? What criteria did they use? Were the grants reviewed individually? Were underlying applications read? Were grantee-specific compliance findings made? Were statutory criteria applied? Did the lawful decisionmaker exercise independent judgment? Or were termination decisions generated through political classifications, AI outputs, and DOGE pressure?
Those are not peripheral questions. They are the questions.
Administrative law is built around a disciplined relationship between record, reason, and action. An agency must explain itself at the time it acts. It cannot rely on post hoc rationalizations after litigation begins. It cannot manufacture clean explanations after the record reveals a different process. It cannot ask a court to defer to reasons the actual decisionmakers did not rely upon.
The NEH terminations exposed the difference between a stated rationale and an actual rationale. The termination notices asserted broad claims about agency needs, priorities, fiscal interests, and the president’s agenda. But the record that later emerged showed keyword searches, DEI classifications, ChatGPT rationales, DOGE-generated lists, and DOGE rejection of NEH recommendations. That mismatch is precisely what administrative law is designed to detect.
A lawful administrative process would have looked very different.
It would have begun with statutory authority. It would have identified the lawful decisionmaker. It would have defined the criteria. It would have reviewed the grant file, not merely a spreadsheet summary. It would have distinguished future policy shifts from post-award terminations. It would have determined whether the recipient violated grant terms or failed to satisfy grant purposes. It would have made grant-specific findings. It would have explained the termination in a way that allowed the recipient, the court, and the public to understand the basis for the action.
That is not what the record described.
The court summarized three facts that drove the legal analysis. First, NEH historically awarded and administered grants through a statutory process centered on expert review, Council involvement, and final action by the chairperson. Second, the April 2025 terminations were not made through any comparable grant-specific process and were not based on recipient-specific findings of noncompliance. Third, DOGE personnel did not merely advise NEH; they overtook post-award administration, directed the termination process, and exercised decisive control using Biden-era status, DEI classifications, keyword searches, and ChatGPT-generated rationales.
That is the administrative-law failure in one paragraph.
The absence of individualized review is especially important. More than 1,400 grants were terminated. But mass government action still requires lawful reasoning. Scale does not excuse process. If anything, scale increases the need for process because error multiplies across every affected recipient. A mistaken classification in one grant file injures one recipient. A defective mass-screening system injures an entire field.
The government’s method also created pretext risk. Pretext is not limited to employment discrimination. In administrative law, courts are alert to situations where the official explanation does not match the real reason. If the public explanation is efficiency, but the record shows viewpoint targeting, identity-linked screening, and unauthorized decision-making, the public explanation cannot simply be accepted at face value. The court must look at the record actually driving the decision.
That is why the ChatGPT and keyword evidence mattered so much. It helped reveal the operative machinery beneath the official language. The government was not merely reviewing grants for budgetary efficiency. The record reflected a process for identifying grants associated with DEI, DEIA, environmental justice, gender ideology, and disfavored subject matter. That process became relevant to First Amendment claims, equal-protection claims, ultra vires claims, and administrative-law review.
The administrative-law lesson is not anti-technology. Agencies can use technology. They can use spreadsheets. They can use search tools. They can use artificial intelligence for appropriate support functions. But lawful tools require lawful purposes and lawful oversight. Technology cannot supply statutory authority. It cannot create a record that the decisionmaker never reviewed. It cannot replace individualized findings where individualized findings are required. It cannot convert political suspicion into reasoned explanation.
The court’s permanent injunction preserved that distinction. It did not bar NEH from administering its grant programs, enforcing lawful grant conditions, or taking future action on lawful and individualized grounds. It simply prevented the government from giving effect to the mass termination adjudicated unlawful and from repeating the same or substantially similar impermissible bases.
That remedy is strategically important. It shows that the court was not freezing the agency in place. It was restoring lawful administration. NEH remained free to do what agencies may lawfully do: administer grants, enforce conditions, and act prospectively within statutory and constitutional limits. What it could not do was enforce a defective mass termination born from unauthorized actors, undisclosed criteria, ideological sorting, and a deficient record.
For the commentary, this section should land on a simple principle: administrative law is not a technical obstacle to efficient government. It is the discipline that prevents efficient lawlessness. When government acts at scale, courts must ask whether the scale reflects lawful administration or mass evasion. Here, the record showed the danger of mass evasion: grants terminated without individualized review, statutory authority displaced, real criteria initially omitted from the record, and public explanations unable to carry the weight of what actually occurred.
That is why the administrative-law claim belongs beside the First Amendment and equal-protection claims. The constitutional violations explain what rights were injured. The ultra vires ruling explains who lacked authority. The administrative-law defects explain how the machinery failed. Together, they show a government process that did not merely make bad decisions. It made decisions in a way the law does not permit.
VIII. The Backfire: Anti-DEI Enforcement Created the Record That Defeated It
The government’s anti-DEI enforcement campaign did not collapse because the court misunderstood the administration’s policy goals. It collapsed because the record showed how those goals were operationalized. That is the backfire at the center of the consolidated NEH grant-termination actions.
The administration could have made a narrower argument. It could have claimed that some federally funded programs engaged in unlawful preferencing, exceeded statutory limits, or failed to comply with grant terms. It could have directed agencies to review future funding priorities within lawful boundaries. It could have required grant-specific analysis. It could have insisted that federal funds be spent consistently with statutes, regulations, and constitutional limits. Those arguments would not automatically prevail, but they would at least sound in law.
That is not the process the record revealed.
The record revealed a mass termination process built around politically loaded labels, keyword searches, abbreviated grant descriptions, ChatGPT-generated “DEI rationales,” subjective categories such as “Craziest Grants” and “Other Bad Grants,” and the displacement of the statutory decisionmaker by DOGE personnel. The very machinery designed to identify and eliminate “DEI” became the evidentiary machinery showing viewpoint targeting, protected-characteristic risk, arbitrary decision-making, and ultra vires action.
That is why the case backfired.
The government’s problem was not merely that it opposed diversity, equity, and inclusion. The problem was that it allegedly used opposition to DEI as an operational filter for terminating grants involving race, gender, religion, Indigenous culture, immigration, public health, women’s history, minority newspapers, identity, and disfavored scholarship. That filter did not produce legal distinctions. It produced constitutional risk.
The keyword evidence showed the problem in concrete form. Terms such as “Tribal,” “Immigrants,” “Indigenous,” “Native,” “Equity,” “Equality,” “Marginalized,” “BIPOC,” “Social Justice,” and “Gay” were not evidence of grant misconduct. They were not evidence of statutory violation. They were not evidence that a recipient failed to perform. Many were simply descriptors of lawful humanities scholarship. But once those terms became detection codes, they helped show that the government was not merely looking for illegality. It was looking for politically disfavored subjects.
That is the evidentiary turn. A slogan became a search protocol. A search protocol became a termination list. A termination list became proof.
This is why the court’s discussion of ChatGPT matters so much. The artificial-intelligence tool did not independently create the constitutional violation. But its use exposed the weakness of the human decision-making process. DOGE personnel asked ChatGPT whether abbreviated project descriptions related “at all” to DEI. They did not define DEI. They did not provide the underlying grant applications. They did not ask whether the grantee violated a statute. They did not ask whether the grantee violated a grant condition. They did not ask whether the project fell outside NEH’s statutory purposes. The tool was asked a politically broad question and generated short rationales that were then incorporated into the termination machinery.
That record was devastating because it showed that the government’s asserted legal judgment may have been nothing more than automated political association. In legal terms, that is not reasoned decision-making. It is classification without discipline.
The examples made the defect impossible to sanitize. A project involving ancient Jewish texts could be characterized as DEI because it concerned Jewish thought. A project on Uyghur persecution could be pulled into the frame because it involved an ethnic and religious minority. A project on American women in Paris could be treated as DEI because it concerned women’s cultural and intellectual history. A project on African-American newspapers could be flagged because it preserved underrepresented voices. These examples did not help the government prove lawful anti-discrimination enforcement. They helped the plaintiffs show that the DEI label was being used so broadly that it captured protected identity, protected subject matter, and protected expression.
That is how the label defeated itself.
The government needed DEI to function as a legal category. The record showed it functioning as a political dragnet. The government needed the court to see the termination process as an efficiency review. The record showed ideological sorting. The government needed DOGE to look advisory. The record showed operational control. The government needed the termination notices to look like NEH action. The record showed DOGE drafting and sending notices under the acting chair’s name.
That mismatch between official form and operational reality is where litigation does its most important work. Public narratives can survive on labels. Court records cannot. Once discovery, depositions, emails, spreadsheets, and internal communications become part of the record, the rhetoric must confront the machinery. Here, the machinery did not support the government’s defense. It undermined it.
The pressure evidence sharpened the point. The record described DOGE personnel pressing for a rapid “game plan,” invoking pressure, and urging NEH officials to remain “on our side.” The court concluded that this was not the language of advisers awaiting a statutory decisionmaker’s independent judgment. That matters because intent and authority often reveal themselves in tone. A lawful adviser offers analysis. A shadow decisionmaker demands alignment.
The government also suffered from a record problem. The court found that the original administrative record did not include the materials necessary to evaluate how the grants were actually selected for termination, including keyword screening, automated processes, generative artificial-intelligence tools, and actors outside NEH. Later supplementation revealed materials showing the criteria and methods used, including ChatGPT and outside actors. That omission reinforced the central concern: the public-facing explanation did not match the actual decision-making architecture.
That is another backfire. An incomplete record may appear useful in the short term because it conceals damaging facts. But once exposed, it becomes evidence of unreliability. It invites the court to ask what else was omitted, why the original record was incomplete, and whether the official explanation was constructed after the fact rather than contemporaneously applied.
The mass scale of the terminations also intensified the problem. A single mistaken termination can sometimes be defended as error. More than 1,400 terminations based on a broad screening system suggest design. The scale made the criteria matter more, not less. If the criteria were unlawful, the error was systemic. If the decisionmaker lacked authority, the defect infected the entire process. If the label was overbroad, the overbreadth multiplied across the humanities field.
That is why the court’s final relief was so consequential. The court declared the mass termination unlawful, unconstitutional, ultra vires, and without legal effect, and permanently enjoined the government from enforcing or giving effect to it or terminating the grants on the same or substantially similar impermissible bases. The remedy recognized that the problem was not an isolated mistake. It was a defective termination regime.
The lesson for public institutions is direct: if the government wants to dismantle unlawful discrimination, it must identify unlawful discrimination. It cannot substitute political suspicion for legal proof. It cannot use identity-linked keywords as proxies for illegality. It cannot use artificial intelligence to generate thin rationales for decisions already politically desired. It cannot bypass statutory decisionmakers and then ask a court to defer to the paperwork.
The anti-DEI campaign backfired because it forgot the basic rule of constitutional government: labels do not decide cases. Records do.
Once the record showed how the terminations were selected, who selected them, what tools were used, what criteria were applied, and how NEH’s statutory process was displaced, the government’s defense became harder to sustain. The government did not simply lose an argument about policy. It lost control of the factual narrative. The anti-DEI apparatus created the very proof that the terminations were not lawful administration.
That is why this ruling should be studied carefully by agencies, lawyers, civil-rights advocates, and courts. The danger is not confined to DEI. Any political label can be used this way. “Extremism,” “waste,” “indoctrination,” “public safety,” “efficiency,” “merit,” “equity,” and “neutrality” can all become enforcement masks if government actors use them without lawful criteria, proper authority, and individualized analysis.
The consolidated NEH actions show what happens when the mask comes off.
IX. The Broader Civil-Rights Lesson: Government Cannot Use Ideology to Evade Equality Law
The broader civil-rights lesson is not that diversity, equity, and inclusion must always be defended. It is not that anti-DEI policies are always unlawful. It is not that courts should manage every policy dispute involving race, gender, identity, scholarship, or public funding. That would overstate the ruling and weaken the analysis.
The lesson is more precise: government cannot use ideology to evade equality law.
Civil-rights law is often misunderstood as a set of rules that belongs to one political side. That is false. Civil-rights law restrains government power. It restrains government when officials use race or sex to exclude. It restrains government when officials use remedial language to impose unlawful preferences. It restrains government when officials suppress speech because they dislike its viewpoint. And it restrains government when officials use anti-DEI language to target protected identity, protected speech, or disfavored cultural memory.
That is why these consolidated actions matter beyond NEH.
The ruling shows that government cannot cure one alleged civil-rights problem by creating another. If an administration believes that federal programs contain unlawful racial preferences, compelled ideological conformity, or discriminatory conditions, it can address those issues. But it must do so through law. It must identify the unlawful conduct. It must define the standard. It must apply that standard consistently. It must use the proper decisionmaker. It must make decisions based on evidence, not slogans. It must distinguish unlawful discrimination from lawful scholarship about discrimination.
That distinction is essential.
A project about Black newspapers is not a racial preference. A project about Indigenous culture is not automatically unlawful DEI. A project about women’s intellectual history is not gender discrimination. A project about Uyghur persecution is not unlawful identity politics. A project about Jewish texts is not improper religious favoritism. These subjects may involve identity, but identity is not illegality.
That is where anti-DEI enforcement can become discriminatory. It begins by claiming to oppose discrimination. It becomes discrimination when it treats the presence of protected groups, marginalized histories, minority cultures, or gendered experience as suspect. It begins by attacking unlawful preference. It becomes unlawful when it attacks visibility, scholarship, memory, and expression.
Civil-rights law is built to recognize that distinction.
This is also why the consolidated posture of the actions matters. The ACLS plaintiffs represented the institutional side of the injury: the harm to scholarly organizations and their members. The Authors Guild plaintiffs represented the individual side: the writers and scholars whose grants were allegedly terminated and whose projects were disrupted. Together, the actions showed how government discrimination can operate both structurally and personally. A policy can damage a field while also injuring individual people.
That duality is familiar in civil-rights litigation. Discriminatory systems often appear abstract until the individual harm is named. The system affects institutions, but it lands on people. Here, the alleged system was a mass termination regime. The individual consequences included lost funding, interrupted scholarship, foregone opportunities, and chilled expression.
The ruling also offers a warning about proxy discrimination. Modern discriminatory action rarely announces itself in old language. Government officials are unlikely to say openly that they are targeting a race, religion, sex, national origin, or protected viewpoint. More often, discrimination travels through proxies: keywords, categories, risk scores, algorithms, funding screens, “efficiency” metrics, or ideological tags.
That is why courts and lawyers must examine function, not just form.
A category may appear neutral on its face but operate as a proxy for protected characteristics. A review may be described as fiscal but operate as ideological punishment. A technology tool may appear objective but apply a biased human prompt. A termination notice may appear official but conceal who actually made the decision. A government record may appear complete but omit the materials that reveal the real criteria.
These are not theoretical concerns. They are precisely the dangers reflected in the NEH record.
The use of technology makes this civil-rights issue more urgent. Artificial intelligence can make discriminatory systems faster, broader, and harder to detect. It can generate rationales. It can classify records. It can produce the appearance of administrative neutrality. But civil-rights law cannot allow technology to become a laundering device for unlawful human intent. If a biased question is automated, the bias does not disappear. It scales.
That is why the ChatGPT evidence belongs at the center of the commentary, not at the margins. It shows how future civil-rights violations may be built: not through a single discriminatory memo, but through prompts, spreadsheets, filters, classification systems, and automated explanations. The discriminatory act may be buried inside the architecture of the decision-making process.
Lawyers must be prepared to find it.
That means asking different questions in discovery. What prompt was used? Who wrote it? What data was supplied? What data was withheld? What definitions were given? What outputs were accepted or rejected? Who reviewed them? Were affected persons allowed to respond? Did the lawful decisionmaker independently evaluate the record? Was the technology used before or after the decision was effectively made? Was it used to analyze, or to justify?
Those questions will matter in employment cases, education cases, grant cases, policing cases, benefits cases, immigration cases, and public contracting cases. The NEH ruling is not only about humanities grants. It is about the next generation of government decision-making.
The broader civil-rights lesson also reaches political rhetoric. The phrase “DEI” has become so politically charged that it often obscures more than it clarifies. Some programs labeled DEI may raise legitimate legal concerns. Some may not. Some may involve unlawful preferences. Some may involve basic access, disability accommodation, historical research, cultural preservation, or anti-discrimination compliance. A serious legal system cannot treat all of that as one thing.
That is why legal precision matters. The law does not ask whether a program has an unpopular label. It asks what the program does. It asks who is affected. It asks what authority supports the action. It asks what criteria were used. It asks whether protected rights were burdened. It asks whether the government acted for a permissible reason.
The government’s failure in these consolidated actions was the failure of precision. It treated DEI as if the label answered all necessary legal questions. It did not. The label only raised more questions.
For civil-rights lawyers, that is the real value of the ruling. It provides a framework for resisting government action that weaponizes ideological terms to target protected people, protected speech, or protected subjects. It also provides a caution against defending every program by invoking moral language rather than legal authority. Both sides of the DEI debate need to hear the same message: slogans do not satisfy constitutional scrutiny.
The Constitution is not impressed by branding.
If a program discriminates unlawfully, it can be challenged. If an anti-DEI enforcement campaign discriminates unlawfully, it can be challenged. The legal system is not supposed to ask which side has the better slogan. It is supposed to ask whether the government acted lawfully.
That is the civil-rights principle worth defending.
Government cannot use inclusion as a mask for unlawful exclusion. Government cannot use anti-inclusion as a mask for unlawful targeting. Government cannot use efficiency as a mask for statutory evasion. Government cannot use technology as a mask for discrimination. And government cannot use ideology as a substitute for law.
The consolidated NEH actions sit at the intersection of all those warnings. They show how quickly a political project can become a constitutional violation when public power is directed at disfavored ideas, disfavored identities, and disfavored scholarship. They also show why courts remain essential in an era when government action can be scaled through technology before affected people even understand how they were selected.
That is the larger meaning of the ruling. The courts did not save DEI. They enforced the boundary between lawful government and ideological punishment.
X. Conclusion: Labels Do Not Override Law
The consolidated NEH grant-termination actions should not be remembered as a simple victory for DEI or a simple defeat for anti-DEI politics. That reading is too shallow. The ruling is more important than the slogan war surrounding it. It is about lawful power.
Government can change direction. Government can review programs. Government can reduce spending. Government can enforce grant conditions. Government can reject unlawful discrimination. Government can decide that future funding should follow different priorities, provided those priorities fit the governing statute and the Constitution.
But government cannot use a label as a substitute for law.
That is what the consolidated actions exposed. The administration’s anti-DEI campaign may have begun as a political objective, but the legal problem arose when that objective was converted into operational machinery: keyword searches, spreadsheet classifications, subjective categories, ChatGPT-generated rationales, DOGE-controlled termination lists, and mass cancellation of previously awarded grants. The machinery did not merely implement policy. It allegedly bypassed statutory authority, targeted protected expression, implicated protected characteristics, and produced a record the court could not reconcile with lawful administration.
That is why the ruling matters.
It tells government actors that “DEI” is not a magic word. It does not dissolve statutory limits. It does not suspend the First Amendment. It does not erase equal protection. It does not authorize shadow decisionmakers. It does not permit artificial intelligence to manufacture reasons for actions already politically desired. It does not allow a federal agency to terminate more than 1,400 grants without individualized, lawful, and reviewable justification.
The same principle would apply in reverse. If government used the language of diversity, equity, inclusion, or accessibility to impose unlawful preferences, punish dissent, exclude disfavored groups, or condition benefits on ideological conformity, that too would raise constitutional problems. The law does not belong to the label. The law governs the conduct.
That is the discipline missing from much of the public debate. DEI is treated as either virtue or vice. Anti-DEI is treated as either restoration or retaliation. But courts cannot decide cases that way. Courts ask what the government did, who did it, what authority permitted it, what criteria were used, what rights were burdened, and whether the record supports the explanation.
In the consolidated NEH actions, those questions exposed the defect.
The statutory structure placed NEH grant authority inside a congressionally designed process. The First Amendment protected scholarship, writing, publication, programming, and expressive work from viewpoint-based government punishment. The equal-protection component of the Fifth Amendment restrained federal action targeting protected characteristics or their proxies. Administrative law required reasoned decision-making, a complete record, lawful criteria, and a real connection between facts and action. Ultra vires doctrine prevented actors without statutory authority from exercising power Congress vested elsewhere.
The mass termination failed across those boundaries.
That is the final lesson: constitutional government is not efficient in the way political movements want it to be. It requires authority before action. It requires process before punishment. It requires evidence before deprivation. It requires lawful criteria before classification. It requires independent judgment before signature. It requires courts willing to look past labels and examine machinery.
That is not bureaucracy. That is the rule of law.
The danger exposed by the NEH ruling is not limited to one administration, one agency, one acronym, or one ideological dispute. Any administration can abuse labels. Any administration can weaponize funding. Any administration can use technology to accelerate improper targeting. Any administration can try to convert political narratives into official action. The safeguard is not faith in the label. The safeguard is law.
That is why lawyers, courts, journalists, scholars, and the public must remain attentive to how government uses language. A term that begins as political rhetoric can become a screening category. A screening category can become a termination list. A termination list can become a constitutional injury. Once government builds machinery around a label, the question is no longer what the label means in public debate. The question is what the label does in official action.
Here, according to the court, it did too much. It reached speech. It reached identity. It reached scholarship. It reached statutory authority. It reached beyond lawful power.
The courts forced the machinery to stop.
That is not a defense of every DEI program. It is a defense of constitutional limits. It is a defense of lawful administration. It is a defense of the principle that government cannot punish disfavored ideas or identity-linked subjects by renaming that punishment “efficiency.” It is a defense of the basic rule that public power must be exercised by the officials Congress authorized, through the process Congress required, for reasons the Constitution permits.
Labels do not override law.
They never have. They never should.
About the Author
Eric Sanders is the founder and president of The Sanders Firm, P.C., a New York-based law firm focused on civil rights, immigration, employment discrimination, police misconduct, and other high-stakes matters. A retired NYPD officer, he brings a rare inside perspective to the intersection of government power, public institutions, enforcement discretion, and constitutional accountability.
Over more than twenty years, Eric has counseled thousands of clients and handled complex matters involving police use of force, sexual harassment, retaliation, systemic discrimination, immigration consequences, and related civil-rights violations. His immigration practice focuses on family petitions, green cards, citizenship, removal defense, humanitarian protection, waivers, appeals, and complex status issues. He graduated with high honors from Adelphi University and earned his Juris Doctor from St. John’s University School of Law. He is licensed to practice in New York State and in the United States District Courts for the Eastern, Northern, and Southern Districts of New York.
Eric has received the You Can Go to College Committee Foundation Humanitarian Award, The Culvert Chronicles 2016 Man of the Year Award, the NAACP—New York Branch Dr. Benjamin L. Hooks “Keeper of the Flame” Award, and the St. John’s University School of Law BLSA Alumni Service Award. He is widely recognized as a leading New York civil-rights attorney and a prominent voice on evidence-based policing, institutional accountability, equal justice, and rights-based immigration advocacy.
