A Brown University economics professor says a spring 2026 class gave him a blunt lesson in what generative AI cheating can look like when assessment moves out of the room.
Roberto Serrano, a blind economist who teaches Brown’s ECON 1170, told El País and Inside Higher Ed that he changed the course’s exam format after a deadly shooting on Brown’s campus in December 2025. Shaken by the attack, which killed two people, including someone who had recently introduced herself to him, Serrano allowed students to take the midterm and final at home.
The enrollment pattern changed fast. El País reported that the difficult course had usually drawn no more than 30 students, and sometimes as few as eight. In spring 2026, 86 students signed up.
Then came the midterm. El País reported an average score of 96 out of 100, with 40 students earning perfect scores. Serrano told Inside Higher Ed that past midterm averages in the course had typically landed between 65 and 80 percent, and that this exam was tougher than earlier ones because students had unlimited time.
The numbers were only part of the problem. Serrano said many answers had an odd, overly elaborate quality. He and graduate students then put the exam questions into ChatGPT and saw responses that resembled what students had submitted, according to the accounts in El País and Inside Higher Ed. That is not a forensic proof machine, no matter how many vendors pretend otherwise. It was enough, though, for Serrano to change the experiment.
He told the class the final would be held in person. In an email quoted by Inside Higher Ed, Serrano said he would count the midterm only if the final’s score distribution looked roughly similar. If not, he said, he expected to void the midterm and weight the final instead.
The class thinned out. El País reported that 18 students dropped the course and nine more did not sit for the final. Of those 27 students, 22 had received 100 on the midterm. Among students who took the in-person final, the average fell to 48.
Brown is already studying the AI problem
The episode lands as Brown is trying to write rules for a tool many students already use. A provost-led Brown report on generative AI in teaching and learning said 56 percent of undergraduate respondents and 67 percent of graduate and medical student respondents reported using generative AI tools daily or weekly.
The same report said large majorities of students expressed worries about how generative AI may affect their learning and cognitive capacity. That detail matters: students are not only asking whether they can get away with using the machines. Many appear to suspect the bargain is bad.
Serrano has made the case in harsher terms. He told Inside Higher Ed that society cannot afford for many of its strongest students to decide cheating is acceptable, calling that a route toward decline. “We cannot choose to become idiots,” he said.
That is a professor’s moral argument, not a statistical audit of Brown’s campus. The confirmed facts are narrower and uglier: a take-home exam produced historically abnormal results, chatbot comparisons raised alarms, an in-person retest produced a collapse, and dozens of students who had excelled on paper disappeared before the room got quiet.
This story draws on original reporting from Ars Technica.
