Writing Grant Politics Into the Rules

When two DOGE staffers sat down to review humanities grants earlier this year, they did not read the applications. Justin Fox and Nathan Cavanaugh fed them into ChatGPT and asked the AI to flag anything that matched a list of keywords: "LGBTQ," "BIPOC," "gender fluidity." The tool recommended terminating 97% of the National Endowment for the Humanities' grant portfolio — more than 1,400 grants worth over $100 million. Cavanaugh later admitted the effort did not reduce the deficit. ^[1] Fox's reasoning for one termination was blunt.

It’s a gender-based story that’s inherently discriminatory to focus on this specific group. — Justin Fox

When a federal judge blocked the mass cancellations in May, she ruled they were unconstitutional viewpoint discrimination. And she added a line that captured something essential about the method.

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. What mattered was that the grant concerned a ’minority group’ — Colleen McMahon

The AI keyword screening was the administration's most deniable approach to reshaping federal grants — and the one that most plainly failed in court. But it was not the only method, and it was not the first. Over six months, the administration has pursued at least three distinct ways of bending federal spending to political priorities, and as each was blocked, the next proved harder for courts to reach. While DOGE staffers were feeding grant abstracts into ChatGPT, the administration was already executing a more direct strategy. In January, Judge Amit Mehta ruled that the Department of Energy had illegally canceled $7.6 billion in clean energy grants. His opinion quoted the government's own admission.

Defendants freely admit that they made grant-termination decisions primarily — if not exclusively — based on whether the awardee resided in a state whose citizens voted for President Trump in 2024. — Amit Mehta

The administration had spared 340 grants in states Trump won or where a Republican senator held sway, while canceling 284 in blue states. This was not a keyword screen producing a plausibly deniable result. It was a political test applied openly. Mehta found it violated the Equal Protection Clause. Then, this week, the administration's own lawyers made the logic explicit: in a court filing, DOE Principal Deputy General Counsel Jeff Novak confirmed the $7.5 billion in cancellations were based solely on the political identity of recipient states. ^[2]

DOE accepts that the inclusion of grants in the October notice tranche was based solely on the political identity of the grant recipient’s state, i.e., whether the recipient’s location and/or place of performance was in a Blue State or a non-Blue State. — Jeff Novak

OMB Director Russell Vought had already supplied the public-facing rationale.

that we have democratic control of the spending that is going out. — Russell Vought

Two methods, two court losses. The keyword screen was blocked as viewpoint discrimination. The voting-record test was blocked as an equal protection violation. So the administration is now trying a third approach — one that does not require hiding behind an AI or admitting to a political litmus test. It is changing the rules. On May 29, the Office of Management and Budget published a proposed revision to the Uniform Guidance, the regulatory framework that governs roughly $1 trillion in annual federal grants. The change would give political appointees formal authority to review and override agency grant decisions to ensure they align with presidential priorities. It would also impose stricter federal approval requirements for publications and prohibit funding research involving specific foreign countries. ^[3] The proposal does not mention keywords or voting patterns. It does not need to. It creates a lawful channel for the same political control that courts found unconstitutional when exercised through ad hoc screens and state-by-state targeting. What was done covertly with ChatGPT and overtly with a blue-state test is now being written into the regulatory code. The arc has not gone unnoticed in Congress. Senate Appropriations Chair Susan Collins, a Republican from Maine, has introduced spending guardrails she describes as without precedent. ^[4]

We never had the kinds of guardrails that we have now. — Susan Collins

The administration's grant politics have moved from deniable to direct to durable. The first method relied on a tool the government could blame. The second relied on a test the government was willing to admit. The third relies on a rule the government is writing itself — one that would make political control of federal spending not a workaround courts can strike down, but the system itself.