Professor suspected AI-powered cheating on take-home midterms, makes finals in-person — only two students scored within 10% of their midterm score
TL;DR
Brown professor Roberto Serrano suspected AI-enabled cheating in his Welfare Economics and Social Choice Theory course after a take-home midterm averaged 96 percent, far above the usual 65 to 80 percent range. He moved the final exam back in person. Of 86 enrolled students, 18 dropped the course and nine did not show up for the final. Among the 59 students who took the final, three scored zero. Only two landed within 10 percent of their midterm score, and just one scored higher than on the midterm.
Nauti's Take
This is the uncomfortable part of the AI-in-education debate: when an exam only measures submitted output, the classroom turns into an API problem. Serrano’s response was harsh, but the score collapse is hard to wave away.
It is fair only if universities pair it with clear rules, real due process, and better assessment formats. Otherwise honest students pay for a system that no longer tests what it claims to test.
Briefingshow
The case shows how quickly take-home exams can break when AI use is poorly defined and hard to prove. The issue is not only cheating, but scale: one professor is expected to document dozens of suspected cases while AI detectors remain unreliable. Universities need assessment designs and integrity processes built for mass incidents, not one-off plagiarism disputes.
Sources