DOGE Put Blackness on the Watchlist

The real scandal is not one sloppy deposition. It is a system that tagged Black life as DEI and called the result neutral.

I was half-listening at first, the way you do when you think you already know the ending.

A deposition video, another political scandal, another grown-up institution being handled like a group project nobody prepared for. I have sat through depositions before, back when I was a cop, and I know the rhythm of them: the slow question, the lawyer circling, the pause where somebody hopes language will hide what the facts are about to drag into the light. I had the tab open, the volume low, and my own quiet arrogance sitting next to the keyboard like a paperweight. Then I heard it. Not a slur. Not a scream. Something colder.

“It is not for the benefit of humankind,” the witness said, explaining why a project about anti-Black violence belonged in the “DEI” bucket. [1]

I felt my face tighten. Not because I believed he meant what it sounded like he meant, even he later called it a “misphrasing.”
I felt it because I know this move. I have heard it dressed up a hundred ways: Black history as a special interest, Black pain as a niche hobby, Black scholarship as an indulgence that polite society can “pause” when budgets get tight.

What set me off was not just the sentence. It was the system behind it. A system that can erase without ever admitting it is erasing.

TLDR

• Newly released discovery in NEH-related lawsuits documents a process where a DOGE staffer used a keyword “Detection Codes” list, including “BIPOC,” “Homosexual,” “Tribal,” and “Gay,” to surface grants for scrutiny, while “white,” “heterosexual,” and “Caucasian” were not on the list. [2][1]

• Under oath, the same staffer confirmed the 120-character ChatGPT prompt used to label grant descriptions as “DEI” and admitted he did nothing to ensure the tool would not discriminate. [1][4]

• The deposition transcript shows a blunt, revealing logic: a documentary about the Colfax Massacre was treated as “DEI” because it centered Black civil rights, with the witness initially claiming that focus was “not for the benefit of humankind,” then walking it back. [1][4]

• Plaintiffs’ filings argue these terminations turned protected characteristics into the sorting mechanism for federal support, raising First Amendment and equal-protection questions while bypassing normal NEH peer-review and termination norms. [3][4]

Restack it and share it. Send it to one friend who still thinks this is “just DEI drama.” If you value this work, support it here:

One-time support:

How we got here

NEH Logo.

This story needs its front porch.

The NEH, the National Endowment for the Humanities, is the federal agency Congress created in 1965 to fund humanities work across the country: museums, libraries, archives, documentaries, preservation projects, public history, scholarship, and community programming. So this is not some niche campus office. It is one of the main ways the federal government helps pay for the country to remember itself. [3]

DOGE, the Department of Government Efficiency, was created by Trump at the start of his second term as the vehicle for his “efficiency” agenda inside federal agencies. Around that same opening stretch, Trump also signed the anti-DEI executive order that told agencies to identify and terminate DEI-related offices, contracts, and grants. That is the bridge. DOGE was the muscle. The anti-DEI order was the mission. [6][5]

Once those two things were in place, the question inside agencies like NEH became brutally simple: which grants can be tagged as DEI, equity, or politically out of bounds for the new administration? According to the court record, DOGE’s small-agencies team got involved with NEH leadership in March 2025, pushed for speed, and helped drive the review and termination of more than a thousand grants by early April. [3][7]

That is what led to the deposition.

Humanities groups and grant recipients sued after the NEH terminations, arguing that the cuts were unconstitutional, unlawful, and driven by viewpoint and identity discrimination. Once those lawsuits entered discovery, plaintiffs got access to emails, spreadsheets, internal records, and sworn testimony from DOGE and NEH officials. Justin Fox was deposed because plaintiffs were trying to answer the most important question in the whole case: who actually made these decisions, by what method, and under what authority? [3][4]

So the deposition is not the beginning of the movie. It is the moment in the movie when somebody finally gets forced to explain the plot under oath.

The thing that wouldn’t leave me alone

Let me separate three things, cleanly, because this story tries to blur them.

Documented fact: A DOGE staffer, Justin Fox, testified under oath that he used a large language model and a keyword approach in his NEH-related work, and the record includes the prompt: “Does the following relate at all to DEI? Respond factually in less than 120 characters. Begin with ‘Yes.’ or ‘No.’…” [1][3]

Strong inference: When your “DEI” detection system is built from identity-coded words, you are not just identifying “programs.” You are identifying people and histories by proxy.

Interpretation: This is what modern erasure looks like when it wants plausible deniability. It does not burn books. It filters budgets. It puts a “Yes” next to your existence and calls it waste. [1][2]

That “benefit of humankind” line mattered because it exposed the inner fantasy behind the machinery: that “humankind” is a neutral category, while Blackness is a deviation from it. The transcript shows the witness later clarifying he was not saying Black history is not for humanity, but the reflex was already out on the table. [1]

And I’ll offer one hesitation line, because it’s honest: maybe I am reading too much into one deposition exchange.

But the evidence does not live in one exchange. It lives in spreadsheets. It lives in the architecture of what got searched for, what did not, and how quickly the “results” became consequences. [1][2]

What everybody is saying

Over the last week, this story broke containment. It moved from court filings to “watch this” clips, to mainstream write-ups, to the kind of pop-political coverage that reaches people who do not read federal dockets for fun. That is one reason it is trending now. [8][9]

The surface narrative goes like this: an “efficiency” operation used ChatGPT to tag “DEI” grants, and a staffer looked unprepared and awkward under questioning. That is true as far as it goes. [1][8]

But the deeper story is not “a guy flailed in a deposition.” The deeper story is procedural. It is about how the state can turn culture war into administrative routine.

The federal executive order at the center of this ecosystem, Executive Order 14151, directs agencies to terminate DEI-related offices and “equity-related” grants and to provide OMB a list of federal grantees who received funding to “provide or advance DEI” or “environmental justice” programs since January 20, 2021. [5]

If you are looking for the bridge between politics and bureaucracy, that is it. The order pulls a moral lever, then hands the machinery a mandate.

And DOGE’s own establishing order frames its purpose as modernizing technology to maximize efficiency and productivity, building teams across agencies, and even setting a defined timeline for its temporary organization ending July 4, 2026. [6]

So when people ask, “How could this happen?” the answer is: it did not drop from the sky. It was a product choice inside a policy atmosphere. [5][6]

The receipts

Here is what the record shows, in plain terms, without the theater.

First, the legal filings describe what NEH normally is supposed to be: a peer-reviewed grantmaking institution with termination historically limited to narrow situations like noncompliance or major deviation from approved objectives. Plaintiffs state NEH has “never before simultaneously terminated hundreds of grants.” [3]

Second, the filings show how the “efficiency” layer arrived. Plaintiffs assert Fox and Nathan Cavanaugh were contacted and hired by senior members of DOGE shortly after inauguration, and that neither had prior government experience relevant to humanities grant administration. [3]

Third, and this is the part that matters for your suspicion about erasure, the filings and exhibits document a two-part sorting method:

1. A pre-meeting keyword scan of NEH grant descriptions using terms like “gay,” “BIPOC (Black, Indigenous, People of Color),” “indigenous,” “tribal,” “melting pot,” and “equality.” [3]

2. A “Detection Codes” list that includes identity and marginalization terms such as “LGBTQ,” “Homosexual,” “Tribal,” “Indigenous,” “BIPOC,” “Privilege,” “Marginalized,” and “Social Justice.” [2][3]

What is not on that list matters as much as what is. The deposition exchange is painfully direct: “white” and “Caucasian” are not on the list, even while “gay,” “LGBTQ plus,” and “homosexual” were. [1][2]

Fox tries to complicate that on the stand. He describes the list as “preemptive” and says it was not used “in determining which grants to keep or not,” at least not in the way the questioner frames it. [1]

That, right there, is the complicating fact that keeps a serious person honest. The defense does not have to be, “We never used a list.” It can be: “We used it as a jumping-off point.”

But then the record shows what happened next: a ChatGPT workflow to label grants. The oath-bound prompt appears in deposition and in the exhibits. [1][3]

And the deposition transcript includes questions about whether he did anything to ensure ChatGPT’s application of “DEI” would not discriminate. The answer, in the filings and summarized in plaintiffs’ memo, is effectively “No,” paired with a repeated insistence that it “didn’t matter.” [1][4]

Now the specific Black-history trigger:

The plaintiffs’ memo describes a grant about the Colfax Massacre, noting ChatGPT classified it as “DEI” because it explored an event that significantly impacted Black civil rights. [4]
In the deposition transcript, Fox agrees with that assessment, then reasons that focusing on a “singular race” makes it DEI, and he initially frames it as “not for the benefit of humankind” before correcting himself. [1]

If you are trying to understand how erasure can wear a startup badge and call itself efficiency, that exchange is a tutorial.

Finally, context from reporting: Inside Higher Ed reports that discovery materials revealed a rushed process, a termination total above $100 million, and that the prompt surfaced in that discovery. It reports that 97 percent of NEH grants were terminated in the sweep described. [7]

The machine under the story

Now we get to the mechanism. Not the vibe. The machine.

This is what I think the machine is, based on what is documented and what is psychologically predictable.

It is a three-part laundering system:

First, policy creates a moral category called “DEI,” broad and flexible enough to hold almost anything. EO 14151 links “DEI” to “waste,” “shameful discrimination,” and mandates termination of “equity-related” grants, plus the compilation of lists of grantees tied to these categories. [5]

Second, the bureaucracy operationalizes that category through proxies. Not by reading projects deeply, but by searching for identity language that can stand in for the category. That is what a detection-code list is. It is not a definition. It is a shortcut with moral force. [2]

Third, the institution outsources the ambiguity to a machine, then uses the machine’s output to create a buffer against accountability. “The model said yes.” The prompt is designed to force a binary. The output is designed to be short. The nuance is designed to be absent. [1][3]

This is not just bad governance. It is a psychological maneuver: what I would call moral outsourcing. You do not have to wrestle with whether defunding a project about anti-Black violence is wrong if the process has already decided that “Black” equals “DEI,” and “DEI” equals illegitimate. You just execute.

NIST warns, in broad terms, that managing AI risks involves attention to negative impacts like threats to civil liberties and rights, and that human organizations can mistakenly assume AI systems “work,” treating outputs with undue confidence. It also emphasizes governance, accountability, documentation, and culture as part of responsible use. [11]

This matters because even if you believe a staffer’s claim that the LLM was not the “ultimate decision-making factor,” a tool used to triage and pre-label 1,162 grants is not neutral. It shapes what gets attention, what gets suspicion, and what is easier to cancel without thinking. [3][7]

And here is the specific racialized double standard the record points toward: it is not that “race” was avoided. It is that “majority” markers were not included as triggers. That asymmetry is not a side detail. It is the shape of the discrimination claim. [1][2]

The deeper incentive underneath all of this is speed. Inside Higher Ed describes pressure to move quickly, and that the sweep happened with startling rapidity in the documented timeline. [7]

Speed loves proxies. Proxies love stereotypes. Stereotypes love cover.

The echo and the bill

Let me say the quiet part out loud: you do not have to announce “I am erasing Black people” to build a system that erases Black people.

That is the historical echo.

Haitian historian Michel-Rolph Trouillot argued that power shapes what is remembered and what is forgotten, what is recorded and what is left out. He framed “silences” as part of how history gets made, not an accident that happens after. [14]

A keyword list is a modern silence tool. A 120-character prompt is a modern silence tool. A spreadsheet labeling projects “crazy” is a modern silence tool. [2][14]

Now, let’s bring it back to the question you asked: was DOGE trying to erase Black folk?

What I can say, responsibly, based on the record: the documented method treats Blackness, and the study of Black life, as a cue for “DEI,” which in this context becomes a cue for cancellation. The transcript shows the witness explicitly agreeing that a project about Black civil rights is “DEI” because it is about Black civil rights. [1][4]

Intent is harder. Courts are built to infer intent from patterns, policies, and outcomes, not from psychic access. What we do have is a set of choices that create foreseeable outcomes.

We also have human consequences.

NEH is not just “academia.” It is libraries, state councils, archives, public humanities programming, community projects, and independent scholars. Inside Higher Ed describes terminations affecting more than $100 million and sweeping most grants, alongside layoffs. [3][7]

And for Black communities specifically, NEH has programs explicitly supporting faculty research at Historically Black Colleges and Universities. That is not rhetorical. It is on the agency’s own site. [13]

When the funding architecture shakes, the first people to lose “optional” resources are often the people who have historically been told their scholarship is “extra.” The record even shows NEH staff flagging grants as “high” or “medium” DEI risk precisely because they addressed racial discrimination and the experiences of Black women, and those grants were later terminated. [3][4]

This is where my humility beat belongs. I have spent enough years listening to people justify harm to know that most harm is not delivered by villains who feel like villains. It is delivered by people who believe they are complying, cleaning up, and moving on.

A word for the people in charge

To the policymakers who built a framework where identity words magically become “waste”: spare me the merit sermon. If you cannot tell the difference between discriminatory preferencing and the study of actual human beings, you are not defending excellence. You are running a forgetting machine.

To the administrators who signed off: what exactly did you think a 120-character yes-or-no chatbot prompt was going to do to complex humanities research? Produce nuance? And what did you think a “Detection Codes” list would do when it flagged “BIPOC” but not “white”? [1][2]

And to the rest of us, the citizens who fund and inherit the record: if you let “efficiency” become the polite word for deleting memory, do not act surprised when the archive starts sounding cleaner, whiter, and more convenient than the country ever was.

Here is the sight line I want you to hold: the future will not call this “a chatbot mistake.” It will call it a method.

And here is the human move: pick one NEH-supported project in your town, your state, your community, an archive, a museum, a humanities council program, an HBCU scholar, and ask what disappears when “DEI” becomes a kill switch. Then tell someone what you learned. Start local. Because erasure always starts local, too.

If this moved you

If that line about Black history being treated as something separate from humanity hit you in the chest, do not just nod and keep scrolling. Restack this. Share it. Send it to one person still sitting on the fence while the record gets cleaned in broad daylight.

Because that is the bet behind all of this. That you will feel the heat, say damn, and move on. Step off the fence.

Paid subscriptions buy the time it takes to read the filings, watch the testimony, and trace the machinery before somebody calls the erasure a misunderstanding:

If paid is not doable right now, you can still back the work here:

---

Sources

1. Exhibit 1: Deposition transcript of Justin Fox (Jan. 28, 2026) - Primary source transcript with the “benefit of humankind” exchange, the ChatGPT prompt discussion, and admissions about screening methods.

2. Exhibit 9: “Detection List” / keyword spreadsheet (filed Mar. 6, 2026) - Primary exhibit showing the “Detection Codes” terms used to flag grants, including identity and marginalization cues like “BIPOC,” “Homosexual,” and “Tribal.”

3. Plaintiffs’ Statement of Undisputed Material Facts (Doc. 249) - Primary court filing that lays out the timeline, normal NEH processes, the keyword strategy, and the ChatGPT workflow as plaintiffs allege it occurred.

4. Plaintiffs’ Memo of Law in Support of Summary Judgment (Doc. 247) - Primary legal argument connecting the evidence to First Amendment and equal-protection claims and giving concrete examples, including the Colfax Massacre grant.

5. Executive Order 14151 in the Federal Register - Primary policy document directing the termination of “equity-related” grants and the compilation of grantee lists tied to DEI and “environmental justice.”

6. White House: Executive Order establishing the Department of Government Efficiency - Primary document defining DOGE’s structure and cross-agency “DOGE Teams,” showing how the “efficiency” mandate is operationalized.

7. Inside Higher Ed: “How DOGE Gutted the NEH in 22 Days” (Mar. 11, 2026) - Reporting synthesis of discovery materials, including scale, timing, and internal-email details about how terminations unfolded.

8. 404 Media: “I Watched 6 Hours of DOGE Bro Testimony…” (Mar. 12, 2026) - Reporting and analysis highlighting the deposition content and why it is circulating widely now.

9. People: “DOGE Staffer Tasked with Eliminating ‘DEI’ Grants Struggles to Define DEI…” (Mar. 12, 2026) - Mainstream coverage showing the story’s broader reach and summarizing key deposition moments for general audiences.

10. The Authors Guild: Brief detailing DOGE’s role in “unlawful” NEH terminations (Mar. 10, 2026) - Plaintiffs’ public-facing summary of claims, emphasizing the ChatGPT and keyword methodology and constitutional theories.

11. NIST: Artificial Intelligence Risk Management Framework (AI RMF 1.0) - Authoritative framework explaining why governance, accountability, and bias risk management matter when AI is used in consequential decisions.

12. Bender, Gebru et al. (2021): “On the Dangers of Stochastic Parrots” - Scholarly paper detailing how large language models can encode bias and why humans over-trust synthetic language, relevant to using LLMs for classification.

13. NEH: Awards for Faculty at Historically Black Colleges and Universities - Agency documentation showing a specific pathway of federal humanities support tied directly to Black institutions and scholars.

14. National Humanities Center: “Silencing the Past: Power and the Production of History” - High-quality interpretive source on how power determines what gets recorded, remembered, and forgotten, clarifying the “erasure” mechanism beyond rhetoric.