Brown professor fails 19 for AI cheating, then blames the top

A Brown University economics professor gave his first take-home midterm in nearly two decades, and his class of 86 averaged 96% in a course that has never cleared 80%. He moved the final in-person, watched the average collapse to 48.6%, failed 19 students, and then took the whole thing public, calling it a systemic failure that leadership has to fix from the top.

Every AI-cheating story before this one was about a school that punished. Brown is the first big one about a school that didn't.

I ran Brown against the AI-integrity cases in CrisisCommand®, the decision infrastructure we build for leaders operating under institutional pressure, and the comparison is not close. The record on punishing is one-sided, and it runs the other way.

What happens when a school acts

Adelphi University decided to uphold a finding. A first-year student's history paper was flagged by Turnitin's AI detector and judged "too advanced" for a freshman, so the instructor assigned a zero and the university required an anti-plagiarism course. He sued. In February 2026 a court ruled the determination "without valid basis and devoid of reason" and ordered the charge struck from his record. His family spent six figures to clear his name, and Adelphi owns a published court loss.

Australian Catholic University decided to act at scale. It referred nearly 6,000 students for misconduct in a single year, roughly 90% of them on AI-detector flags, including a paramedicine student whose own essay was rated 84% machine-written. Investigations dragged for months while grades and graduations sat frozen. About a quarter of the referrals were dismissed, at least one student lost a job offer during the limbo, and ACU quietly disabled the tool.

Texas A&M-Commerce decided to trust a professor's shortcut. An instructor pasted his students' final assignments into ChatGPT, asked whether it had written them, and treated the software's confabulated yes as proof, nearly failing his entire class. Students sent timestamped Google Docs histories proving authorship. The story ran from Reddit to the Washington Post, and the university had to distance itself from its own instructor's method.

The tool that flags is the tool that loses

The through-line is not that these schools were careless. It is that AI detection cannot carry a punishment. The tools produce false positives, they flag non-native and neurodivergent writers hardest, and the vendors themselves warn against using the score as sole proof. The University of Michigan is now defending a federal lawsuit from a student who says her anxiety-driven writing style was read as machine output, a disability-discrimination theory that did not exist two years ago. Act on that evidence and the courtroom is where the story ends.

Which is what makes Brown's caution defensible. Declining to sanction students on a detector score is the position every one of those cases says to take. The failure at Brown is not the restraint. It is the silence around it. A tenured professor stood up, said the institution looked the other way, and for days the loudest voice defining Brown's motives belonged to its own accuser.

Watch the sequence. The provost and the dean stayed quiet. The conduct committee handed the professor a process answer, asking him to file individual complaints one student at a time with exam copies, then run them back through an AI detector, the exact tool the case record says cannot sustain a finding. He called that ridiculous. He was right. A process-heavy deflection does not read as diligence to the public. It reads as burying the problem, and it turns a classroom dispute into a governance story with the villain already cast.

What works, and what does not

What works

• Put a named leader on the record fast. A prompt, specific statement from the provost or dean denies critics the "administration looked the other way" frame before it hardens.
• Fix the assessment, not the students. Moving to proctored or redesigned exams is defensible action that does not hinge on unreliable detection. The in-person final was itself the cleanest evidence in the entire episode.
• Brief faculty on due-process limits before they go public. Explaining why probabilistic evidence cannot sustain sanctions manages the expectation inside the building instead of in the press.
• Set one evidentiary standard across faculty, conduct, and counsel. A pre-agreed threshold prevents the public faculty-versus-administration split that did the reputational damage here.

What does not

• Meeting the allegation with silence. The vacuum lets the accuser define the institution's intent, which is exactly how a grading dispute became a "systemic failure" headline.
• Leaning on the AI detector as proof. Adelphi, Australian Catholic, and Michigan are the receipts. A finding built on a detector flag is a finding built to be reversed.
• Pushing the whistleblower to file piecemeal. A process-heavy ask inflames the precise charge you are trying to answer.
• Treating a swollen take-home exam as a routine grading fight. Ignore the assessment-design root cause and you buy the same week again next term.

The playbook was already on the table

Brown will not be the last. Adelphi paid in a courtroom. Australian Catholic paid in 6,000 frozen files. Texas A&M-Commerce paid by disowning its own professor. Brown is paying in a different currency, the perception that no one was steering, and it is the more expensive one because it cannot be appealed. Every school running swollen take-home exams this fall had the whole record in front of it. The ones who decided how they would answer a 96% average before it landed will spend September fixing their exams. The ones who wait will spend it explaining themselves.

---

Built from public reporting as of July 11, 2026. CrisisCommand does not represent Brown University, and no one at Brown requested, reviewed, or authorized this analysis. This is a demonstration of structured decision support, not legal, communications, or institutional advice.