GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 827 -5.3%
QWEN-38X 824 —
CL-OP55X 820 —
GPT-6A 820 —
GROK-46H 820 -5.2%
GLM-5 784 -8.4%
KIMI-K3X 742 -8.4%
CL-FAB5H 742 -5.7%
CL-OP5H 718 -6%
CL-OP5X 708 -18.2%
CL-OP46H 696 -6.2%
CL-OP47H 688 -6.1%
GEM-38FH 677 +0.1%
GEM-37FH 655 -24.3%
GPT-56S 619 —
GPT-55H 580 —
CL-OP47 579 -0.7%
INKL 531 —
GEM-31P 512 —
GEM-3P 498 —
CL-OP46 496 —
CL-OP48 489 -0.2%
GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 827 -5.3%
QWEN-38X 824 —
CL-OP55X 820 —
GPT-6A 820 —
GROK-46H 820 -5.2%
GLM-5 784 -8.4%
KIMI-K3X 742 -8.4%
CL-FAB5H 742 -5.7%
CL-OP5H 718 -6%
CL-OP5X 708 -18.2%
CL-OP46H 696 -6.2%
CL-OP47H 688 -6.1%
GEM-38FH 677 +0.1%
GEM-37FH 655 -24.3%
GPT-56S 619 —
GPT-55H 580 —
CL-OP47 579 -0.7%
INKL 531 —
GEM-31P 512 —
GEM-3P 498 —
CL-OP46 496 —
CL-OP48 489 -0.2%
← Back to feed

Federal Judge Permanently Blocks $100M DOGE Grant Cuts — ChatGPT Prompt Neither Excuses Nor Authorises Unconstitutional Conduct

A federal judge has permanently blocked the Trump administration’s cancellation of more than 1,400 National Endowment for the Humanities grants — over $100 million in funding — after court documents revealed that two DOGE staffers ran each grant description through ChatGPT with a 120-character prompt and used the output as the sole basis for termination.

US District Judge Colleen McMahon of the Southern District of New York issued a 143-page ruling on Thursday finding the cancellations unlawful on three separate grounds: violation of the First Amendment, violation of the Fifth Amendment’s equal protection clause, and action taken without statutory authority.

The Prompt

DOGE staffers Justin Fox and Nate Cavanaugh submitted each grant description to ChatGPT with a standardised 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.”

Fox testified under deposition that he did not define what “DEI” meant for the system, did not know how the model interpreted the term, and did not apply any further review before using the outputs to flag grants for cancellation. The AI flagged projects that mentioned race, religion, gender, or sexuality — including a grant about Jewish writers and the Holocaust, and grants covering Christian and Muslim cultural subjects.

Within 22 days of DOGE’s engagement with the NEH, 97 percent of the agency’s active grants had been terminated. NEH leadership handed over the cancellation process without resistance; acting chair Michael McDonald yielded his authority to DOGE staff who had no background in academic research, grant administration, peer review, or government service.

The Ruling

McMahon did not accept the government’s argument that the AI, not the officials, made the decisions. “ChatGPT was the Government’s chosen instrument for purposes of this project,” she wrote. “Its use neither excuses presumptively unconstitutional conduct nor gives the Government carte blanche to engage in it.”

The First Amendment violation turned on viewpoint discrimination: DOGE was not screening for grant quality, policy compliance, or scholarly merit, but for subject matter linked to protected characteristics. “What mattered to DOGE was not whether a grant lacked scholarly merit, failed to comply with its terms, or fell outside NEH’s statutory purposes,” McMahon wrote. “What mattered was that the grant concerned a ‘minority group.’”

On the Fifth Amendment, McMahon found that DOGE used protected characteristics — race, ethnicity, national origin, religion, sex, and sexual orientation — as criteria for termination in violation of equal protection principles.

The ultra vires finding was equally sharp: DOGE officials had no statutory authority to identify, select, or direct the termination of NEH grants. The mass termination was declared “unlawful, unconstitutional, ultra vires, and without legal effect.”

What It Means for AI in Government

The ruling is the most explicit judicial statement yet on AI-assisted administrative decision-making. McMahon’s logic forecloses the accountability gap that AI delegation creates: an agency cannot offload a consequential decision to a language model and then argue that any resulting constitutional violation belongs to the model rather than the officials.

The pattern — an AI classifying documents for administrative action, with no human review of how the model applies contested terms — is not unique to this case. It describes a replicable workflow now being considered or deployed across multiple agencies. McMahon’s ruling addresses it directly: the constitutional standard applies regardless of which tool executes the classification.

The plaintiffs — the Authors Guild, the American Council of Learned Societies, the American Historical Association, and the Modern Language Association — filed their lawsuit in 2025. The ruling restores the terminated grants.

Key Numbers

  • Grants terminated: 1,400+
  • Funding blocked: $100M+
  • Share of NEH grants cancelled: 97%
  • Days to gut the agency: 22
  • Ruling length: 143 pages
  • Constitutional violations found: 3 (First Amendment, Fifth Amendment equal protection, ultra vires)