US federal agencies searched active research grants by keyword before terminating funding
US federal agencies searched active research grants by keyword before terminating funding
Agreements with federal agencies submitted with a court motion describe a common process: agencies first identified grants by searching for specific words and topics, then selected some of those grants for termination. In some cases, artificial intelligence tools were used to compile the lists.
On July 15, the plaintiffs in Thakur v. Trump filed a motion in federal court and attached agreements with the agencies. The National Science Foundation, the National Endowment for the Humanities, the US Departments of Defense and Transportation, and the National Institutes of Health described how they selected grants for possible termination.
“Search terms, keywords, or phrases” were used to flag grants for review.
According to the agreements, agencies then selected projects from these lists if the projects expressed, or were believed to express, positions that the administration did not support. The agencies applied broad criteria and used standardized letters instead of determining separately whether each grant recipient had complied with the funding terms. The same agreements state that the recipients had not violated those terms.
At the National Institutes of Health and the Department of Health and Human Services, the range of topics expanded from projects concerning diversity, equity, and inclusion to gender, vaccine hesitancy, and COVID-19. The lists included “health equity,” “structural racism,” and “sexual orientation.” The materials do not explain what data the AI tools processed or how their outputs affected the decisions. For this agency, the documents state only that AI may have helped prepare the lists submitted to it.
On July 21, the Associated Press reported that the case concerns more than a thousand University of California grants. The plaintiffs estimate that the funding previously awarded totaled approximately $2 billion. A hearing on the motion is scheduled for October 20, and the court has not yet decided whether this procedure is lawful.
This differs from the White House plan to test rules for awarding new grants experimentally, which proposes comparing methods for selecting applications with the resulting research outcomes. The agreements in Thakur describe a review of grants that had already been awarded, based on topic related words and the position a project was believed to express.
Continuity matters for long-term biomedical research: grant funding supports the hiring of research teams, the recruitment of participants, and experiments that continue for years. The materials do not establish the consequences for each laboratory. However, the process they describe creates a risk for such programs because funding for research already underway may be reconsidered without a separate assessment of whether the laboratory complied with the grant terms. This applies to aging research as one area of long-term biomedical research, although the documents do not identify any specific geroscience programs.