Brown professor says AI cheating followed take-home exam

Brown University economics professor Roberto Serrano says a take-home midterm he offered after a deadly campus shooting revealed widespread AI-assisted cheating. The case has become a sharp example of how generative AI is forcing colleges to rethink exams, honor codes and academic discipline.

Serrano told Fortune he changed the format of a March 5 exam in his ECON 1170 course, an advanced undergraduate class in mathematical economics, because students were still shaken by the Dec. 13 attack at Brown. Fortune reported that two of his students were among nine people wounded, and that both survived after weeks in critical condition.

One of the fatal victims, Ella Cook, had recently met with Serrano about becoming her academic adviser, he told Fortune. In response to the trauma on campus, Serrano gave students a closed-book take-home midterm, a format he said he had not used in 34 years at Brown.

Unusual scores raised alarms

The results stood out. Fortune reported that 40 of the 86 students who took the midterm received perfect scores, and that the class average was 96. Serrano said prior averages in the course had generally fallen between 65 and 80, even though he had made this exam harder than usual.

Serrano told Fortune that graders later compared student answers with responses generated by ChatGPT. He said several exams used an unusually complex line of reasoning that matched the AI output, even though the problem had a simpler proof.

El País had previously reported on the cheating allegations. Serrano told Fortune the grade pattern and repeated reasoning made the evidence of fraud overwhelming.

Rather than cancel the midterm, Serrano told students the final exam would be held in person. He said that if the final’s results did not resemble the midterm’s distribution, only the final would count toward grades.

Drops and failures followed

After Serrano addressed the class about what he had found, 27 students dropped the course, according to Fortune. Of those students, 22 had received 100 on the take-home midterm.

Fortune reported that 59 students sat for the in-person final and 19 failed. The final exam average fell to 48 out of 100, which Serrano told Fortune was the lowest final average in the course’s history.

Serrano said he sent his findings to Brown’s dean of the college and provost. He told Fortune he did not initially receive responses, and that after he brought the matter to the Academic Code Committee, he received a note describing the episode as a “wake-up call.” Fortune reported that Brown had not responded to its requests for comment.

Serrano holds the Harrison S. Kravis University Professorship in Economics at Brown and edits the journal Games and Economic Behavior, according to Fortune. He is also the author of widely used economics textbooks and received the King of Spain Prize for Economics in 2024, Fortune reported.

Colleges reassess AI and exams

Serrano told Fortune he believes universities have been slow to deal openly with AI’s effect on academic integrity. He said take-home exams are no longer viable in his course because students can too easily use AI systems to complete the work.

For the next academic year, Serrano told Fortune, weekly homework will carry no weight in final grades, and take-home exams will not return. He said AI may have useful educational roles, but colleges need clearer limits and consequences when academic integrity is breached.

Other universities are also changing policies. Fortune cited Princeton faculty’s May vote to require proctors for exams, ending a 133-year tradition of unsupervised testing. Fortune also reported that 57% of U.S. college students say they use AI tools in coursework weekly, and that 47% of surveyed Harvard seniors admitted to cheating.

This story draws on original reporting from Fortune.