Brown University Professor Exposes Mass ChatGPT Cheating on Exam

Economics professor Roberto Serrano of Brown University presented evidence that dozens of his students had used ChatGPT to cheat on a take-home exam in advanced mathematical economics. When he switched the exam back to an in-person format, the class average dropped from 96 to 48 out of 100, and nearly a quarter of the group didn't show up to the exam room at all.

Serrano opted for a take-home exam for the first time in his career after a December shooting on Brown University's campus that killed two students, including Ella Cook, whom the professor had recently agreed to advise academically. He wanted to ease the burden on students during an emotionally difficult period. The outcome, however, turned out to be surprising for an entirely different reason.

How the Professor Caught the Cheating

When Serrano noticed that the average score drastically exceeded the norm from previous years, he decided to test his hypothesis of mass cheating. Together with his graduate students, he fed the same exam questions directly into ChatGPT and compared the model's answers with the students' work. Some answers contained identical, unusual phrasing that matched what the chatbot had generated.

Before announcing the results, he gave students a transparent rule: if the final exam results differed significantly from the take-home exam, the take-home would be voided. When he announced that the final would be held in person and without access to AI tools, 18 students immediately dropped the course, and nine more, despite being formally enrolled, didn't show up to the exam. Of those 27 absent students, 22 had scored a perfect 100 on the take-home exam.

Scale of the Problem in the US

The Brown University case is not an isolated one. Similar stories from recent months involve other American universities, including Western University in Canada, where a professor likewise threw out exam results over suspected AI-assisted cheating. A study conducted at Princeton University found that nearly 30 percent of students admit to using artificial intelligence to cheat on exams or homework at least once.

Thanks to AI, the cost of cheating has dropped to essentially zero - Roberto Serrano, economics professor, Brown University

If employees are just going to press a button to ask an AI agent to do their work for them, then we're signing up for a world in which humanity has decided to make itself stupid - Roberto Serrano, economics professor, Brown University

University Response and Professor's Reaction

Serrano makes no secret of his frustration with the lack of a systemic response from university administrators. In his view, academic institutions across the United States were caught off guard by how quickly generative AI upended the realities of testing, and still haven't worked out a coherent policy for dealing with it. The professor says openly that silence is the worst possible response to the problem, and that universities risk losing credibility if they don't start defending academic integrity directly.

The case has sparked wide discussion in American academic media, from Inside Higher Ed to the Chronicle of Higher Education, as one of the most thoroughly documented instances yet of mass chatbot-driven exam cheating at a prestigious Ivy League school.

What It Means for Polish Universities

For Polish universities, the Brown University case is a warning that traditional unsupervised take-home exams and papers have lost credibility as tools for assessing knowledge. Polish universities, like their American counterparts, are only now working out rules for how students can use artificial intelligence, and many lecturers still rely on assessment formats that a chatbot can easily solve. Serrano's story shows that returning to in-person, handwritten exams may today be the only way to reliably verify knowledge in fields that require independent reasoning.

Experts point out that the solution isn't just tighter controls, but also rethinking what and how universities actually want to teach in an era when access to advanced language models is widespread. Serrano himself admits the problem won't disappear just because exams return to the classroom, since students will keep using AI in their everyday studying anyway - the question is how to teach them to do it in a way that builds skills rather than replacing them.

Sources: Fortune (fortune.com), Tom's Hardware (tomshardware.com), Promptowy (promptowy.com)