The Most Controversial Acronym in America Has a Definition.
The People Enforcing It Just Won't Use It.
Here is what happened to American humanities funding in April 2025. A twenty-something named Nate Cavanaugh, freshly arrived from tech and finance with no government or humanities experience, opened ChatGPT, typed prompts built around Trump’s executive order on DEI, and fed 1,477 grant descriptions through it. The ones the model flagged, he recommended for cancellation. Over $100 million in grants were cut. Oral history programs dissolved. Archival research stopped mid-project. Scholars lost years of work.
When deposed in early 2026 as part of a federal lawsuit brought by the American Historical Association, the American Council of Learned Societies, and several other humanities organizations against the National Endowment for the Humanities, Cavanaugh was asked whether he felt remorse that researchers had lost income. He said: “No.”
When asked to define DEI, the term he had spent months enforcing, his colleague Justin Fox pointed to the executive order. DEI was whatever the order described.
Two things are happening in this story and they are not separate. A major federal enforcement action was built on a term no one would define. The instrument of that enforcement was an AI tool. These facts belong together because they produce the same outcome: decisions with enormous human consequences, made by people insulated from those consequences, using systems that cannot see what they are destroying. That is not a coincidence. It is a condition. And it is the condition DEI was specifically designed to diagnose and correct.
Before we go further, we need to define the term.
What DEI Is
Diversity is the presence of difference within a group: race, gender, age, background, ability, experience. It describes who is in the room.
Equity means allocating resources according to what people actually need to reach comparable outcomes, because people do not start from the same place.
Equality means giving everyone the same thing regardless of where they started.
The distinction is concrete. Two people join the same company on the same day. One grew up with college-educated parents, attended a well-funded school, and arrived with an established professional network. The other was the first in her family to finish college, attended an underfunded school, and knows no one in the industry. Identical onboarding is equality. Recognizing that identical resources will produce very different outcomes and designing support accordingly is equity. Same goal. Different method, because the starting conditions differ.
DEI as institutional practice asks whether systems produce what they claim to value. Who advances, at what rate, and why. It is a management discipline with a documented return. Companies in the top quartile for ethnic diversity on executive teams are 39 percent more likely to outperform peers on profitability. Diverse management teams report innovation revenue 19 percent higher than homogeneous competitors. Organizations that embed equity in core strategy, rather than isolating it in a separate office, sustain performance gains that peers do not.¹
This is a performance argument, not a values argument. The reason DEI became a political target is that it produced results consequential enough to disrupt existing distributions of power. That is the whole story.
Now consider what AI systems do when built and deployed by teams that lack diversity. They learn from historical data. They encode the assumptions of whoever built them. They learn that leadership looks like whoever has historically held it, that creditworthy looks like whoever has historically received credit, that fundable scholarship looks like whatever has historically been funded. Cavanaugh’s ChatGPT prompt is not a metaphor for AI bias. It is a direct instance of it: a tool trained on the internet’s existing hierarchies, operated by someone whose professional world never required him to examine those hierarchies, cancelling the work of scholars that world has historically excluded.
DEI and AI belong in the same analysis not because they are both tech-adjacent topics but because they raise the same question: who gets to define normal, and what happens to everyone else when that definition goes unexamined.
Take This Test
This exercise has circulated in American workplaces and universities for thirty years, adapted from Peggy McIntosh’s 1989 essay on invisible advantage.² Mark a 1 if the statement applies to you. Mark a 0 if it does not.
I can arrange to spend most of my time in the company of people of my race.
If I need to move, I can be reasonably confident my new neighbors will treat me neutrally or pleasantly.
I can turn on the television and see people of my race widely and positively represented.
When I am taught my country’s history, people of my race are credited with shaping it.
I can find a hairdresser who knows how to work with my hair, and products for my skin in any drugstore.
I do not need to teach my children to be alert to racism for their own daily safety.
I am not asked to speak for all people of my racial group.
If I ask to speak with the person in charge, I expect to be directed to someone of my race.
If a police officer pulls me over, I am reasonably confident race was not the reason.
I can easily find children’s books, toys, and media featuring people of my race.
I return from work feeling connected, not isolated or out of place.
If I am late, my tardiness is not attributed to my race.
I can accept a job or promotion without colleagues suspecting race was the reason.
I can buy bandages and cosmetics that approximately match my skin.
I can travel with a partner without expecting hostility from the people who serve us.
As the only person of my race in a room, I still expect my perspective to be heard.
I can afford to be inattentive to the experiences of people of other races without professional or social consequence.
I can find mentors of my race willing to invest in my professional advancement.
When my leadership is questioned, I am confident race is not a factor.
I feel normal and welcome in everyday institutional and public life.
Add up your score.
Your score reveals more than you might expect. Read the full piece to understand what it means, the historical pattern behind the current assault on DEI, why the backlash narrative has cause and effect exactly backwards, and what you can actually do about it.
What Your Score Means
A score above 16 means your race has functioned as an asset in American institutions. Doors opened. Systems felt neutral because they were built around people like you.
A score below 10 means your race has been a variable requiring active management, in ways most people who scored above 16 have never had to consider.
Each statement maps to a system. Items 1 and 2 are housing. Item 3 is media. Item 4 is education. Items 6 and 9 are policing. Item 8 is leadership pipelines. Item 13 is hiring and promotion. Item 18 is professional sponsorship. These are institutional conditions with measurable, generational consequences by group. Not feelings. Facts with causes.
Item 6 is worth pausing on. Before a Black or Brown teenager can drive, their parents sit them down and explain precisely how to behave if a police officer stops them. Hands visible. Speak slowly. Say “sir.” Narrate every movement before making it. This is not a conversation about anxiety. It is a safety protocol, passed between generations, because the cost of getting it wrong can be a life. Most white parents have never had this conversation and never will. That difference, replicated across dozens of daily decisions and interactions, compounding across an entire career and lifetime, is what the test is measuring.
Race shapes all American lives, not only Black and Brown ones. The person who scored 20 has a race as certainly as anyone else. It has simply operated as the assumed baseline, so embedded in institutional design that it registers as the absence of race rather than as race itself. A fish does not notice water. It simply swims. The moment you talk honestly to someone swimming in different water, you begin to see what you are both in.
Return to Cavanaugh. He was swimming in his water, using a tool built by people in the same water, making decisions about scholarship produced by people in entirely different water. He felt no remorse because he could not see what he was destroying. DEI was the institutional practice that would have required him to ask, before flagging a single grant, what these projects were, who produced them, and what their elimination would mean. That practice is what he was eliminating. The circularity is the point.
Why This Is Happening
America has run this pattern before, with enough consistency that its structure is documentable.
After Reconstruction, when Black Americans held office, built businesses, and accumulated property, the reversal was organized and comprehensive. Black Codes restricted movement and labor. The Supreme Court’s 1896 decision in Plessy v. Ferguson ratified legal segregation as constitutionally adequate. Progress, then retrenchment, in the language of order.³
After Brown v. Board of Education, Southern states closed public schools rather than integrate them. Prince Edward County, Virginia shuttered its entire school system for five years. Rhetoric: states’ rights. Object: a racial order the courts had ruled unconstitutional.⁴
After the Civil Rights Act, Nixon gave us “law and order.” Reagan gave us the welfare queen. Civil rights enforcement slowed through the 1980s and the representation gains of the previous decade began to reverse.⁵
The structure is consistent across all three periods: meaningful progress, organized backlash framed as a defense of neutral principles, dismantling of the mechanisms that produced the progress. The vocabulary updates. The project does not.
In a November 2025 event at Princeton with Eddie Glaude, Kimberlé Crenshaw, who coined the term intersectionality, described the current moment as a continuation of that project. The goal is not to debate whether DEI works. It is to remove the framework for the debate, so that disparities become background noise again. Outcomes without causes. Facts without examination, because the tools for examination no longer exist.⁶
What is new in this iteration is the infrastructure. Knight Foundation data show that white-male-owned firms manage approximately 99 percent of assets under management in the U.S. asset-management industry.⁷ The venture capital and technology firms that build and deploy AI reflect a similarly narrow demographic. When those systems score job applications, allocate credit, or, as here, determine which scholarship is ideologically permissible, the people most affected are the least represented in the rooms where those decisions are made.
White men make up roughly 30 percent of the U.S. population and hold 62 percent of C-suite positions in Fortune 500 companies. Women represent 51 percent of Americans and 26 percent of executive roles. Black Americans are 13.6 percent of the population and 3.2 percent of executive leadership.⁸ These numbers describe who builds the systems, who enforces them, and who gets to define what counts as normal. When AI codifies normal, it codifies those numbers. That is not an AI problem in isolation. It is a representation problem that AI makes faster, larger, and harder to see.
On the Backlash
The dominant explanation for the current moment goes like this: the left overcorrected after 2020. It became preachy and punitive. DEI was a casualty of overreach.
Look at what that requires you to accept. After the murders of George Floyd and Breonna Taylor, Americans demanded that Black people not be killed by police without consequence. After MeToo, women demanded workplaces where they could not be assaulted by men who controlled their careers. The overreach narrative locates the provocation in those demands. It asks you to conclude that the people being harmed caused the backlash against the systems harming them, by asking too clearly to stop.
That is a reversal of cause and effect and should be named as one.
What is also true is that some tactics deployed after 2020 were counterproductive. Mandatory trainings designed to surface discomfort rather than change systems generated resistance rather than reflection. Publicly shaming individuals rarely produces accountability. It produces defensiveness. Those failures were real and cost real political ground.
A failed tactic is not a failed demand. The demands were legitimate. What fueled the backlash was not how they were made but that they were being answered. As other groups advanced, some white men experienced that advance as loss. That feeling was real. It was also deliberately amplified by political actors who understood that zero-sum thinking is potent precisely because it is intuitive.
It is also wrong. McKinsey Global Institute estimates that closing gender and racial gaps in the American workforce would add trillions to GDP over the coming decade.⁹ The American cities that have thrived in this century attracted the widest range of talent. More for others is not less for you. It is a larger economy in which everyone participates.
Clear the Fog
The Cavanaugh story is not a DEI story with an AI detail. It is an AI story that shows exactly what DEI was for. An undefined term, a ChatGPT prompt, and two people with no stake in the outcome cancelled $100 million in humanities research without being able to say what they were looking for. That is not a policy failure. It is a representation failure: decisions with enormous human consequences, made by people with no knowledge of those consequences, automated at scale. The fog and the algorithm are the same tool. Both work by removing the people most affected from the room where the decisions are made.
A word about “woke.” Using it as a slur, even in jest, even with the eye-roll that signals you know better, is accepting the terms of the con. The con works like this: naming a documented statistical reality, that white men hold a disproportionate share of American money, power, and institutional influence, gets recast as an attack on white men. Designing systems to close that gap becomes reverse discrimination. The person describing the problem becomes the problem.
None of that survives contact with the data. It is a rhetorical trap that has worked because otherwise clear-headed people have stepped into it to seem reasonable.
Do not step into it. The numbers are a measurement, not a political position. The systems that produced them were built by people, which means they can be rebuilt by people. But only by people willing to see them clearly and name them accurately.
What to Do
The litigation matters. Plaintiffs in the NEH case are arguing that the grant cancellations were arbitrary, premised on a definition the enforcers could not articulate under oath. Those deposition videos have been public since March 2026. The ACLU, the NAACP Legal Defense Fund, and PEN America are pursuing related cases and need resources.¹⁰
The data matters. Ask your organization what representation data it has stopped collecting. Ask what AI tools it is using to make hiring, promotion, or evaluation decisions, and who was in the room when those tools were built and tested. The structural failure that produced the Cavanaugh deposition is reproduced every time a consequential AI system is deployed without diverse oversight. These are the same problem with different interfaces.
The vocabulary matters. When “DEI” is used as a slur without a definition, ask for the definition. A federal employee enforcing a policy under oath could not provide one. That fact belongs in every conversation where these words are being weaponized.
People built these systems. The Civil Rights Act, the Americans with Disabilities Act, the legal frameworks that made workplace harassment actionable: none of those came from inevitability. They came from people who refused to accept that outcomes were natural, insisted on naming what was producing them, and built the infrastructure to change it.
Say the words. Name the systems. Refuse the fog.
Please invite others to join as well.
Footnotes
McKinsey, “Diversity Matters Even More,” 2023; Boston Consulting Group, “Inclusion Isn’t Nice, It’s Necessary,” 2023; Harvard Business Review, “The Five Stages of DEI Maturity,” November 2022.
Adapted from Peggy McIntosh, “White Privilege: Unpacking the Invisible Knapsack,” Peace and Freedom Magazine, July/August 1989.
Eric Foner, Reconstruction (1988); Plessy v. Ferguson, 163 U.S. 537 (1896).
Bob Smith, They Closed Their Schools (1965).
Thomas Sugrue, The Origins of the Urban Crisis (1996).
The Daily Princetonian, “Crenshaw and Glaude Reflect on Attacks on DEI,” November 10, 2025.
Knight Foundation, “Diversifying Investments: A Study of Ownership Diversity and Performance in the Asset Management Industry,” 2021.
U.S. Equal Employment Opportunity Commission, “Diversity in High Tech,” 2023; McKinsey, “Women in the Workplace,” 2024; Coqual, “Being Black in Corporate America,” 2023.
McKinsey Global Institute, 2023; Citi GPS, “Closing the Racial Inequality Gaps,” 2020.
ACLU, NAACP Legal Defense Fund, PEN America active litigation dockets, April 2026.
I believe that DEI is the benchmark for what a civilized society needs to strive forward and what we are seeing now is an old desire, being acted upon, to return to the period prior to the Civil War when slavery and prejudice ran amok. It remains to be seen whether the U. S.goes back to the future or gets mired in the past.
Some flies in the ointment, asked not to argue but clarify.
First, you mainly mention race which is difficult inasmuch as scholars agree it doesn’t exist. Also, positing DEI "describes who's in the room" assumes that what matters about those present is their genetically-determined membership in a historically oppressed group. Right?
Second, you argue for race-based DEI on the grounds that equity improves financial outcomes. This appeal to shareholder returns is silly when fairness is the issue at hand. Some of the most appalling systems ever were quite good for shareholders but it doesn't change the fact that fairness matters most.
Third, you ignore the fact that regardless of outcome, DEI relies on literal racism - giving different treatment to groups based on what you presume to be true due to their ethnicity. This linking of advantages ignores the largest actual predictor I believe - the wealth of one’s parents.
DEI is most manifest on college campuses and has been for years. Whether these experiments in social Darwinism have been net-good, as a Brown grad I can tell you it was an uncomfortable exercise in applied hypocrisy to attend a school still named after slave traders, lectured as a straight white man when two generations earlier, my fellow Jews were not admitted. Also being told what literature we could and couldn't read by Park Avenues finest censors while they explained I couldn't be a Feminist, not because of what I believed but because I was a man.
This long comment on your piece is based only on the fact that I think you're arguing for something harmful for the country I love. I'm proud to say I've never participated in a race or gender-based hiring process. When it comes up I gently ask why the most important criteria isn't merit and explain why profiling otherwise is an insult to all involved. And yes, I've hired many members of whatever classification system you advocate but only because they were the best.
I know my position is not popular but it’s based on a belief that different standards for different identities is pernicious. You may assume I think this as a cis white man but it’s actually something I earned with a feminist academic mom, fighting against apartheid and living with a major disability.
Every component of DEI is laudable but in achieving its aims is sacrifices each one of them.