DOGE Used ChatGPT to Cancel $100M in Grants — A Judge Ruled It Both Dumb and Illegal
Photo by Vitaly Gariev on Unsplash
- On May 7–8, 2026, U.S. District Judge Colleen McMahon issued a 143-page ruling declaring DOGE's cancellation of over $100 million in National Endowment for the Humanities grants unconstitutional.
- DOGE used a vague ChatGPT prompt to flag and terminate more than 1,400 grants — including a Jewish literary anthology it mistakenly identified as DEI content — without ever defining what "DEI" meant to the AI.
- The court ruled the process violated the First and Fifth Amendments, and that DOGE had no legal authority to cancel the grants in the first place.
- The case is a real-world warning for any organization about what happens when workflow automation replaces human judgment in high-stakes decisions.
What Happened
In April 2025, the Department of Government Efficiency — widely known as DOGE — terminated more than 1,400 grants awarded by the National Endowment for the Humanities (NEH). These grants supported scholars, research institutions, and cultural organizations across the United States. The total value of the cancelled grants exceeded $100 million. The trigger was an executive order titled "Ending Radical and Wasteful Government DEI Programs and Preferencing."
The method DOGE used to screen those grants was a single ChatGPT 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." The AI was never given a legal or official definition of what "DEI" actually meant. It was essentially guessing — at scale, with public money.
On May 7–8, 2026, U.S. District Judge Colleen McMahon issued a sweeping 143-page ruling declaring these terminations unconstitutional. She found violations of the First Amendment (viewpoint discrimination — meaning the government penalized speech based on its content) and the equal-protection component of the Fifth Amendment. She also ruled that "DOGE had no statutory authority" — meaning no legal power granted by Congress — to cancel NEH grants at all.
The ruling permanently bars the administration from terminating the grants and orders the NEH to reinstate all of them. The White House called the decision "egregiously wrong" and signaled an appeal. One of the ruling's most striking examples: an anthology titled "In the Shadow of the Holocaust: Short Fiction by Jewish Writers from the Soviet Union" was flagged by ChatGPT as DEI content — a Jewish literary history collection misread by an AI with no defined standards to work from.
Photo by Conny Schneider on Unsplash
Why It Matters for Your Team's Productivity
You might be thinking: "That's a government problem, not mine." But the DOGE ChatGPT case is actually a masterclass in how not to use AI automation — and the lessons apply directly to small businesses and remote teams relying on productivity software every day.
Here's the core failure: DOGE gave ChatGPT an ambiguous task with no defined criteria, no human review process, and no accountability loop. Think of it like hiring a new employee to review 1,400 contracts and mark some for cancellation — but never telling them what "problematic" actually means. That's what happened, except the "employee" was an AI, the stakes were over $100 million, and the result was a 143-page court order.
For small business owners, this scenario plays out in miniature all the time. Teams adopt business tools with enthusiasm, then automate decisions they shouldn't: auto-rejecting job applications based on keyword scanning, auto-flagging customer complaints as spam, or routing refund requests through a chatbot with no clear instructions. The technology is rarely the problem — the undefined process almost always is.
Judge McMahon described the grant screening as "a textbook example of unconstitutional viewpoint discrimination." In plain terms: the AI was penalizing ideas it couldn't even understand. For your team, the equivalent risk is a workflow automation that accidentally treats certain customers, employees, or partners unfairly — because nobody wrote down what "fair" means in the system's rules.
The good news is that the best saas tools available today — platforms like Monday.com, Notion, ClickUp, and Zapier — are designed to support human-in-the-loop review (meaning a real person stays involved in final decisions) rather than blindly automating consequential choices. A responsibly designed workflow automation surfaces information for humans to act on; it does not replace human judgment entirely.
For remote teams specifically, team collaboration breaks down fast when automated systems make opaque decisions that no one can explain or defend. If your team can't answer "why did the system do that?" — you have a process problem that no AI feature will solve on its own.
The broader legal context matters too. Federal courts have been increasingly scrutinizing AI-assisted administrative decisions for compliance with due process (the right to a fair, transparent procedure) and the Administrative Procedure Act (a law requiring government agencies to explain and justify their decisions). Legal scholars expect these standards to influence how courts treat private-sector AI decisions in employment, lending, and customer service contexts as well. Whatever productivity software your team uses, the question "can we defend this decision?" needs to have a clear answer.
Photo by Milad Fakurian on Unsplash
The AI Angle
Building on those productivity risks, it's worth examining exactly where the AI process collapsed — and what it signals for how you deploy AI inside your own business tools.
The DOGE prompt was technically functional: ChatGPT answered every query without error. The failure was governance — the rules, oversight, and accountability structures around how the tool was used. No one defined "DEI." No one validated the outputs against a standard. No one reviewed edge cases before acting on them. The result was a Jewish literary anthology misclassified as diversity content — an error any informed human reviewer would have caught in under a minute.
The best saas tools built for serious workflow automation — including AI-enhanced platforms like Notion AI, Make (formerly Integromat), and ServiceNow — include audit logs (permanent records of every automated action), confidence thresholds (settings that flag uncertain outputs for human review), and approval gates (required human sign-offs before consequential actions execute). These features exist precisely to prevent unchecked automation from creating legal or operational disasters. When evaluating any productivity software with AI capabilities, ask vendors directly: "What happens when the AI isn't sure?" That answer tells you more about the tool's real-world reliability than any feature list.
What Should You Do? 3 Action Steps
The DOGE failure started before ChatGPT was ever opened — it started when decision-makers failed to define what they were actually looking for. Before using any workflow automation to screen, filter, or categorize data, write out in plain language the exact criteria the system should apply. If you cannot define it clearly for a human reader, you cannot define it clearly for an AI. Audit your current business tools and flag any automated decision that relies on vague or undefined instructions — then rewrite those instructions before your next review cycle.
AI excels at processing volume; human judgment is still essential for consequential decisions. Review your team collaboration workflows and identify every automated action that could harm a customer, employee, vendor, or partner if it goes wrong. Add a required human approval step before those actions execute. Most leading productivity software platforms — including Monday.com, ClickUp, and Asana — support native approval workflows. If yours does not, that is a meaningful signal to evaluate alternatives.
Judge McMahon's ruling highlighted the Holocaust anthology example precisely because it was an obvious edge-case failure that should have been caught before any action was taken. Set a recurring calendar reminder — monthly or quarterly — to pull a random sample of your AI-assisted decisions and review them manually. Look specifically for outputs that seem surprising, counterintuitive, or inconsistent. If your AI is flagging things that do not make sense, your prompt, your training data, or your criteria need revision. This kind of regular audit is standard practice for teams using the best saas tools responsibly in regulated or customer-facing environments.
Frequently Asked Questions
Is it legal for companies to use AI tools like ChatGPT to make hiring or compliance decisions in 2026?
It depends on how the AI is used and in what jurisdiction. Using AI for preliminary research or summarization is generally permitted, but using it as the sole decision-maker — especially without defined criteria or human oversight — creates significant legal exposure. The DOGE ruling illustrates that courts are increasingly skeptical of AI-only decision processes when they affect people's rights or funding. In employment contexts, EEOC guidelines and several state laws impose additional requirements around automated screening tools. Always consult qualified legal counsel before automating consequential decisions that affect people.
What workflow automation mistakes should small businesses avoid after the DOGE ChatGPT ruling?
The three biggest mistakes to avoid are: (1) automating decisions without writing down clear, specific criteria the AI can actually apply; (2) removing human review from high-stakes or irreversible outcomes; and (3) never auditing what your automated systems are actually deciding in practice. The DOGE case combined all three errors simultaneously — and the price was $100 million in cancelled grants and a 143-page federal court ruling. Small businesses face the same pattern at a smaller scale, but the reputational and legal consequences can still be severe. Treat workflow automation as a research tool, not a replacement for accountable human decision-making.
How do the best SaaS tools handle AI decision-making to reduce legal and compliance risks for teams?
The best saas tools in the compliance and workflow automation space — including platforms like ServiceNow, Notion, and Monday.com — typically include audit trails (permanent, searchable records of every automated action), human-in-the-loop approval gates, and confidence thresholds that escalate uncertain AI outputs for manual review. When evaluating any productivity software with AI capabilities, ask specifically about these governance features. A tool that cannot tell you what it did, why it did it, and who approved it is a liability risk, not a productivity gain.
Can the DOGE NEH grant cancellations still be appealed, and what does that mean for affected organizations in 2026?
Yes. The White House signaled it would appeal Judge McMahon's May 2026 ruling to a higher federal court. However, the permanent injunction — the court order blocking the grant cancellations — remains in effect unless a higher court explicitly suspends it. The NEH is ordered to reinstate the more than 1,400 terminated grants in the meantime. The appeals process typically takes months to years, leaving affected scholars, universities, and humanities organizations in a state of uncertainty even as their grants are technically restored. Organizations dependent on federal grants should consult legal counsel about contingency planning during the appeal period.
What is the difference between AI-assisted and AI-automated decisions, and why does it matter for team collaboration tools?
An AI-assisted decision is one where the AI provides research, summaries, or ranked recommendations — but a human makes and owns the final call. An AI-automated decision is one where the system acts entirely on its own without meaningful human review. The DOGE process was automated: ChatGPT flagged grants, and those grants were terminated without a substantive human check. For your team, the distinction matters because courts, regulators, and customers increasingly hold organizations responsible for automated decisions in ways they do not for decisions where a human was clearly accountable. The safest philosophy for any team collaboration or productivity software deployment: automate the research, not the ruling.
Disclaimer: This article is for informational purposes only and does not constitute legal advice. Tool features, pricing, and legal interpretations may change. Always verify current details on official websites and consult qualified legal counsel for compliance-related questions specific to your organization.
No comments:
Post a Comment