“BIPOC (Black, Indigenous, People of Color),” “Minorities,” “Native,” “Tribal,” “Indigenous,” “Immigrant,” “LGBTQ,” “Homosexual,” and “Gay.”
These were the search terms utilized by the federal government to eliminate Diversity, Equity, and Inclusion (DEI) in federal funding, all while using Artificial Intelligence (AI) to target those funds. In a stunning rebuke of such discriminatory practices, the District Court for the Southern District of New York found in American Council of Learned Societies v. McDonald (ACLS), that the record established “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.” The federal government terminated grants involving African, Asian, and South American countries—as well as grants relating to Indigenous Americans and Jewish women—while retaining grants involving Western European subjects.
Against this backdrop, the DEI legal landscape can be incredibly challenging. From anti-DEI investigative subpoenas to Civil Investigative Demand (CIDs), warnings in the form of “Dear Colleague” letters, and new proposed anti-DEI regulations, recipients of federal funding are easy targets of anti-DEI governmental enforcement.[1] The government’s use of AI raises the stakes even higher.
The Constitutional Line
Clarity can be drawn, however, from recent court decisions that prohibit federal government activity that crosses the legal line regarding constitutional protections for recipients of federal grants, who have been targeted because of perceived connections to DEI initiatives. The federal government cannot lawfully terminate federal funding merely on the basis of protected characteristics, such as race, religion, sex, or association with DEI-related topics. Indeed, the message is clear—the government’s criticism of DEI through its termination of federal funding is unconstitutional. And when the government uses AI in an effort to identify and eliminate DEI programs they do not agree with, the message is equally clear—the government’s use of AI that results in viewpoint discrimination or discrimination against protected groups is unconstitutional.
In Thakur v. Trump, federally funded researchers challenged the government’s decision to terminate their grants, because the funded projects were related to DEI and other viewpoints disfavored by the Trump administration. In its decision, the U.S. Court of Appeals for the Ninth Circuit held that the First Amendment’s prohibition against viewpoint discrimination extends to government funding decisions. Because the grants at issue were terminated based on recipients’ perceived connection with DEI-related topics, the court concluded that the government’s terminations of their funding violated the First Amendment. The Ninth Circuit affirmed the lower court’s decision and ordered the government to reinstate the previously terminated research grants.
AI as a Discriminatory Sword and Shield
The Southern District of New York applied the same reasoning in ACLS, holding that the federal government’s termination of DEI-related humanities grants violated both the First Amendment and the Fifth Amendment’s Equal Protection component. The decision is significant because it sharply criticized the government’s use of generative AI in evaluating grants, describing the process as evidence of arbitrary agency action and ideologically motivated discrimination, despite the government’s attempt to scapegoat ChatGPT as responsible for the use of discriminatory classifications.
The dispute in ACLS arose after the National Endowment for the Humanities (NEH), under direction from the Department of Government Efficiency (DOGE), conducted a review of humanities grants awarded during the Biden administration to determine their connection, if at all, to DEI.[2] DOGE personnel searched through abbreviated descriptions of projects using terms such as “equity,” “tribal,” and “immigrants.” Eventually, DOGE began using ChatGPT to classify grants and to draft rationales for why they might be DEI-related. Critically, “DEI” was never defined for ChatGPT, nor was it understood how the system interpreted the term. As a result, ChatGPT repeatedly classified projects as DEI-related for irrational and arbitrary reasons, often because projects bore mere reference to minority groups, women, immigration, or cultural history. DOGE failed to conduct any review or reject any of the AI-generated classifications. The resulting AI-generated classifications were used by DOGE to justify termination of identified grants.
The district court illustrated this arbitrary process with clear examples of viewpoint discrimination. One terminated project that examined the “dark history” of government-run boarding schools for Native Americans was classified as DEI, because “it concerned the historical mistreatment of Indigenous children by the federal government and sought to educate the public about that injustice.” Another terminated project, which studied the graphic narratives of Jewish women survivors of Nazi persecution in the immediate aftermath of the Holocaust, was deemed DEI-related based on its “Jewish” and “women” areas of focus. A digitization project of African-American newspapers was flagged as DEI-related because it aimed to preserve “underrepresented voices.”[3] Lastly, a historical project concerning American women in Paris was labeled DEI-related, apparently because it focused on women’s experiences and accomplishments.
The district court also found that the federal government violated the Equal Protection clause and identified direct evidence of unconstitutional discrimination based on protected traits. It compared two grants concerning veterans’ experiences: one project focused broadly on veterans, while the other focused specifically on Chinese-American veterans. DOGE declined to terminate the broader veterans project because it merely “focus[ed] on veterans” but terminated the Chinese-American veterans project because it sought to “uplift and recognize the contributions and experiences of Chinese-American veterans.” The only meaningful distinction between these two projects was the Asian-American component of the terminated grant—a race-based distinction forbidden by the Equal Protection component of the Fifth Amendment.
The district court held that terminating grants on the sole basis of their relation to DEI were “textbook examples” of unconstitutional viewpoint discrimination in violation of the First Amendment. Further, choosing to only scrutinize Biden-era grants constituted prohibited political-association discrimination. The court also held that the government violated the Equal Protection Clause by terminating grants based on protected traits, such as gender, race, and sexual orientation.
The Judiciary’s Check on Executive Branch Abuses
These rulings establish a strong and definitive legal boundary: the federal government lacks the constitutional authority to revoke federal funding simply over DEI affiliations or because of protected characteristics, such as race, ethnicity, religion, or sex. Ultimately, this litigation signals that anti-DEI retaliation via funding withdrawal is unconstitutional—and the judiciary may intervene to stop it. Federal government investigations, enforcement and regulatory activity may continue to be subject to legal challenges if they are capriciously and punitively based on protected group discrimination, viewpoint discrimination, or political-association discrimination, with or without the use of AI.
Endnotes
* Kyle Chao, a Summer Associate—not admitted to practice—in Epstein Becker Green’s Los Angeles office, contributed to this article’s preparation.
[1] The Office of Management and Budget (OMB) has proposed that anti-DEI regulations go into effect on October 1, 2026 to “to ensure that only a single set of government-wide requirements apply to Federal awards made during fiscal year 2027.” Comments on the proposed regulations are due on or before July 13, 2026.
[2] As reported in the Washington Post, “[t]he U.S. DOGE Service Temporary Organization [was] terminate[d] on July 4, 2026,” See https://www.washingtonpost.com/politics/2026/07/06/elon-musks-us-doge-service-announces-its-formal-end/.
[3] DOGE also directed the termination of a grant supporting the National Ornamental Metal Museum in Tennessee. The museum’s project focused on the “preservation, promotion, and advancement of the field of fine metalwork.” Nevertheless, the project was categorized as DEI-related because it sought to promote “understanding and inclusivity in the field.”
Blog Editors
Authors
Member of the Firm
Associate