The federal keyword lists that canceled billions in research funding
highereddive.com
August 17, 2026
4 min read
61/100
Summary
Court filings in a lawsuit brought by University of California researchers show that the National Institutes of Health, National Science Foundation, Defense Department and National Endowment for the Humanities used keyword searches and general criteria to identify grants for termination, rather than individually assessing every award. The agencies said grants were cut when projects expressed, or were presumed to express, viewpoints disfavored by the Trump administration. The searches focused on diversity, equity and inclusion, climate, green-energy and related subjects. NSF drew its terms from a 2024 report by Sen. Ted Cruz and searched hundreds of words, including “injustice,” “ally,” “prejudice,” “discrimination,” “minority” and “traumatic.” NIH used terms including “workforce diversity” and “health equity,” plus permutations. The Defense Department searched for terms such as “pay equity,” “LGBTQ,” “neurodiversity,” “climate change,” “decarbonization” and alternative-energy sources including solar, wind and geothermal. NEH flagged terms including “environmental justice,” “transgenderism” and DEI variants. NIH canceled at least $2.6 million in funding for projects that used “structural racism,” among other targeted terms, and canceled additional awards that had diversity supplements. NIH also suspended more than $500 million in UCLA funding, which a court later ordered restored. UC researchers contend the terminations violated the First Amendment, and have obtained two preliminary injunctions blocking form-letter grant cancellations across several agencies.
Key Takeaways
- Four federal agencies told a court that they used keyword lists and general criteria to identify University of California research grants for termination rather than reviewing every grant individually.
- NSF used hundreds of search terms drawn from a 2024 Sen. Ted Cruz report, while NIH, the Defense Department and NEH searched terms associated with DEI, climate policy, green energy and gender identity.
- NIH canceled at least $2.6 million in grants associated with the phrase “structural racism” and also terminated grants that received diversity supplements.
- A court ordered the restoration of more than $500 million in UCLA funding that NIH had suspended, and UC researchers have won two preliminary injunctions against form-letter grant terminations.
What the discussion said
The thread was overwhelmingly furious about the reported grant cancellations, but its AI-specific discussion centered on whether automation helped create the mess. One commenter proposed that an LLM could screen proposals more intelligently than literal keyword matching, recognizing context instead of treating every occurrence of a flagged term as disqualifying. That idea drew a sharp rebuttal: if the goal is to remove politically disfavored research, a language model is not a cure for crude policy but a scalable way to make the same bad judgment. Several readers also suspected that generative AI may have helped produce or operationalize the list itself, pointing to ideological phrasing that sounded less like a considered research policy than a model prompted to mimic partisan internet rhetoric. This was speculation rather than established evidence, but it fit the broader fear that opaque automated triage can turn vague political directives into mass false positives. The strongest shared conclusion was not that the system needs a smarter classifier; it was that no classifier should be deciding scientific support from viewpoint-laden language. There was no meaningful enthusiasm for AI here, aside from the tentative observation that context-aware analysis could technically outperform exact-string filtering.
Where opinion split
Could an LLM have made grant screening less absurd than a fixed blacklist? One side suggested semantic model-based review could avoid punishing harmless uses of terms; opponents argued that automating ideological cuts with an LLM merely upgrades the machinery of arbitrary censorship, and some believed such models were already part of the process.
Community Sentiment
Positives
- Context-aware language models could, in principle, distinguish a proposal’s actual subject from an accidental keyword hit, avoiding the most laughable failures of exact-match filtering.
Concerns
- Replacing literal keyword matching with an LLM would not fix politically driven grant cuts; it could simply automate ideological judgment at far greater scale.
- Speculation that a generative model helped craft the list reflects fear that partisan prompting can launder internet conspiracy language into consequential research policy.
- The false-positive examples make automated text triage look dangerous for research: models or rules that ignore context can erase legitimate scientific work over incidental vocabulary.
