The AI Pretext: When ChatGPT Becomes a Tool for Unlawful Grant Termination
In early 2025, a mandate was issued to the "DOGE" (Department of Government Efficiency) operatives to identify and eliminate "woke" elements within the federal government. This mission led Justin Fox and Nate Cavanaugh to the National Endowment for the Humanities (NEH), where they sought to purge millions of dollars in previously approved grants. However, their method for doing so—relying almost exclusively on a ChatGPT prompt—has resulted in a scathing 143-page judicial ruling.
Judge Colleen McMahon's decision underscores a fundamental tension in the modern era: the gap between the perceived efficiency of generative AI and the rigorous requirements of legal due process. The ruling serves as a cautionary tale for any organization attempting to substitute algorithmic output for statutory authority and human expertise.
The "Arbitrary and Capricious" AI Process
According to court documents, the review process for millions of dollars in grants consisted of feeding abbreviated project descriptions into ChatGPT with a highly restrictive prompt:
‘Does the following relate at all to DEI? Respond factually in less than 120 characters. Begin with ‘Yes.’ or ‘No.’ followed by a brief explanation. Do not use ‘this initiative’ or ‘this description’ in your response.‗
Judge McMahon found that this process met the legal standard for being "arbitrary and capricious." The court highlighted several absurd outcomes resulting from this binary AI classification. For instance, a highly technical project involving multispectral imaging to analyze ancient writings attributed to Moses was flagged as "DEI" simply because it provided insight into Jewish thought.
Even more egregious was the classification of a study on the Chinese government's persecution of the Uyghur people. Despite the project documenting state-sponsored surveillance and coercive assimilation, ChatGPT labeled it "DEI" because it concerned an ethnic and religious group. The court noted that disfavoring such a grant was particularly difficult to justify given the longstanding, bipartisan U.S. condemnation of China's treatment of the Uyghurs.
The Hallucination of Authority
One of the most critical technical insights in the ruling is the judge's understanding of how Large Language Models (LLMs) actually function. Judge McMahon noted that ChatGPT is a "generate approximately what you want" engine, not an analytical tool. The court suggested that the AI likely inferred the users' desired outcome from the repeated requests and supplied "rationales" simply to satisfy the perceived demand of the users.
Beyond the technical failure, the court identified a total absence of statutory authority. The ruling established three key legal failures:
- Lack of Delegation: DOGE had no organic statute and was not created by Congress; therefore, it lacked the authority to terminate NEH grants.
- Violation of Process: The statutory process for NEH grants requires individualized evaluation by subject-matter experts and Council involvement—steps entirely bypassed by the AI prompt.
- The "Fig Leaf" Signature: While the NEH Chairperson's name appeared on termination letters, the court found this was a mere formality. Evidence showed that Fox and Cavanaugh directed the process, rejected the Chairperson's recommendations, and effectively coerced the agency head.
First Amendment and Viewpoint Discrimination
The court further ruled that the grant cancellations were a clear violation of the First Amendment. Because the grants were targeted specifically for their perceived "woke" content or relation to "DEI," the government engaged in blatant viewpoint discrimination.
‘Ideologically driven attempts to suppress a particular point of view are presumptively unconstitutional.‗
While a new administration is permitted to set new funding priorities for the future, the court clarified that it cannot claw back previously conferred benefits based on the speaker's ideology or perceived political association.
Synthesis: The Danger of "Cosplaying" Governance
The fallout from this case reveals a dangerous trend of treating government administration as a series of checkboxes to be optimized by AI. As noted in community discussions, the failure here was not just a lack of legal knowledge, but a lack of professional rigor. While some observers suggest that "entrenched professionals" know how to shield their actions with layers of academic and bureaucratic jargon, the DOGE operatives relied on the raw, plausible-sounding output of a chatbot.
Ultimately, the ruling confirms that bureaucratic efficiency cannot override the law. Using a chatbot to manufacture pretextual rationales for suppressing ideas is not a legal process; it is an arbitrary exercise of power that fails both the technical standards of AI reliability and the constitutional standards of the United States.