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 moved the final exam back in person after unusually high take-home midterm scores raised suspicions that many students had used AI tools. Out of 86 enrolled students, 18 dropped the class after the announcement and nine skipped the final. That left 59 students taking the in-person exam. Only two students scored within 10% of their midterm result, and only one improved on the midterm. Three students scored zero.
Nauti's Take
This does not look like a small anomaly; it looks like a warning sign arriving on schedule. Take-home exams with vague AI rules are no longer a robust assessment format, they are an experiment in incentives.
The better response is not simply to drag every exam back into the classroom, but to separate what may be built with AI from what students must explain, defend, or solve live. Otherwise, the gray area rewards nerve more than competence.
Briefingshow
The case shows how quickly take-home exams can lose credibility in the age of AI. The issue is not only cheating, but assessment design: if a task can be outsourced to ChatGPT, it may measure tool access, prompting skill, and risk tolerance more than understanding. Universities need exam formats that account for AI use upfront, not only discipline after the fact.
Sources