A Brown professor moved his exam back in person, and the class average dropped from 96 to 48

[TrainingRun Anvil]

Brown University economics professor Roberto Serrano says he uncovered what looks like mass AI-assisted cheating in his own classroom after comparing take-home and in-person exam results from the same students. His Welfare Economics and Social Choice Theory course usually attracts a small cohort, sometimes as few as eight students, topping out around 30 in a typical year. This spring, enrollment jumped to 86

Following the December 2025 mass shooting at Brown that killed two students and injured nine others, Serrano allowed his class to take their midterm from home rather than forcing a return to a supervised classroom so soon afterward. The results immediately raised red flags, the class averaged 96 out of 100, with 40 of the 86 students scoring a perfect or near-perfect grade, even though Serrano says the take-home format let him write a harder exam than usual given students had unlimited time. In previous semesters, the class average had ranged between 65 and 80

Suspecting widespread AI use, Serrano moved the final exam back into a supervised, in-person setting. Of the 59 students who sat that final, only two had scores comparable to their earlier take-home midterm results, the broader pattern showed scores collapsing from an average of 96 down to roughly 48 once AI tools were no longer an option. Y Combinator's Paul Graham shared a chart of the results on social media, noting that based on the visible spread, all but three students appeared to have used AI on the midterm

Serrano was careful to say he doesn't want to demonize AI as a technology, but argues universities need to draw firmer lines around its use in graded assessment. We cannot afford to have a society in which a significant fraction of our best young minds think that cheating is okay, he said. That leads to a declining society, to a failed society. We cannot choose to become idiots. He also criticized Brown's institutional response as meek so far. The case lands alongside a wider reckoning across US higher education, Brown's own Generative AI in Teaching and Learning committee found 56 percent of undergraduates use AI tools daily or weekly, and a 2025 Princeton survey found 27.7 percent of seniors admitted using ChatGPT in unauthorized ways, both numbers rising sharply year over year

Brown Professor Suspects Most of His Class Used AI to Cheatwww.insidehighered.com