Brown University Professor Catches Nearly Entire Class Cheating With ChatGPT

Roberto Serrano, a professor of welfare economics at Brown University, has taught the same course for nearly twenty years. This spring, he swapped the in-class midterm for a take-home exam for the first time, partly because students were wary of sitting in a lecture hall after a shooting on campus in December. The result exceeded his worst suspicions.

Serrano and his teaching assistants ran the exam questions through ChatGPT and compared the model's answers with the students' work. They found passages that matched what the chatbot generated, answers that were technically correct but phrased in a distinctive, overly elaborate style he did not expect from students. Some used unusual proof-by-contradiction arguments instead of the methods taught in class, exactly as ChatGPT had proposed them.

The Professor's Response

After analyzing the results, Serrano sent the class a message stating openly that he suspected most students of cheating. With the dean's approval, he switched the final exam from take-home to in-class. The effect was immediate: some students dropped the course before the final, and the average score among those who remained fell by nearly half compared with the midterm.

The distribution of scores made it clear that something seriously troubling had happened - Roberto Serrano, professor of economics, Brown University

If workers just press a button to have AI do the work for them, then we are writing a world in which humanity has chosen to be idiots - Roberto Serrano, professor of economics, Brown University

University Response Cautious

When Serrano handed the data over to Brown's Academic Code Committee in May, he received no response. Only after he took the case public in the media in late June did the committee ask him to file individual complaints against specific students. Meanwhile, a university-appointed AI committee published recommendations urging faculty to "not emphasize punishment" and to avoid restrictive rules on AI use.

The disconnect, a faculty member sounding the alarm over mass cheating while university administrators recommend a softer approach, shows how far American universities still are from a coherent policy on chatbots. Serrano is not alone in his concerns: a similar share of faculty reported worry in the national AACU survey even before the latest models launched.

What It Means for Education

The Brown case comes at a moment when access to advanced chatbots is growing faster than universities' ability to verify independent work. Take-home exams, long treated as a less stressful alternative to in-class testing, are becoming difficult to defend as a reliable form of assessment when every student carries a tool in their pocket that can solve a welfare economics problem in seconds.

For Polish universities, it is a warning from across the ocean: the problem affects any institution that bases grading on unsupervised work. Some Polish higher education institutions are already introducing their own rules on AI use in coursework, but Serrano's case shows that even prestigious American universities are still searching for answers on how to assess students in a world where access to a capable language model is universal.

Sources: Inside Higher Ed (insidehighered.com), Fortune (fortune.com)