The average score in a Brown University economics class fell from 96 out of 100 on a take-home midterm to 48 on an in-person final after professor Roberto Serrano forced the class into a proctored room, suspecting widespread use of ChatGPT. Eighteen of the 86 enrolled students dropped ECON 1170 before the final, and nine more failed to show up. Of those 27 students, 22 had scored a perfect 100 on the midterm.
Serrano introduced take-home exams for spring 2026 after a December 2025 campus shooting left him shaken. The class, an advanced economics course, had historically capped at 30 students and sometimes ran with just 8. Under the new format, enrollment jumped to 86. The March 5 midterm produced 40 perfect scores against a historical average band of 65 to 80 percent.
The style of the answers, not just the scores, tipped Serrano off. Correct responses had what he described as a convoluted phrasing that felt off. When he and his graduate students ran the exam questions through ChatGPT, the model returned similar answers.
“Historically the average grade in the midterm of this course has ranged between 65 and 80 [percent], and this exam was harder than the exams I wrote in the past, because… take-home is an opportunity to challenge the class a little bit more.”— Roberto Serrano, Brown University economics professor
Key facts
- 01Brown's ECON 1170 midterm averaged 96 out of 100 with 40 perfect scores; the in-person final averaged 48, a 50% drop.
- 0218 students dropped ECON 1170 and 9 skipped the final after Serrano announced an in-person exam — 22 of those 27 had scored a perfect 100 on the midterm.
- 03Enrollment jumped from a historical max of 30 (sometimes just 8) to 86 students after Serrano introduced take-home exams for spring 2026.
- 04A Princeton survey found 29.9% of students admitted to using AI to cheat on at least one exam or assignment.
- 0556% of Brown undergraduates and 67% of graduate and medical students reported intentionally using generative AI tools daily or weekly.
Rather than void the midterm outright, Serrano moved the final in-person and emailed students that he would compare the two distributions before deciding.
The exit was immediate and lopsided. Eighteen students dropped the course after the announcement. Nine more enrolled students did not attend the final. Among those who did sit the exam, the average score collapsed by half.
“I am not declaring [the midterm] void for now. I am going to give the class a chance to prove me wrong. That is, if the distribution of the final exam is roughly similar to the distribution of the midterm, I will count the midterm.”— Roberto Serrano, Brown University economics professor
The Brown incident is a controlled experiment, however unintentionally, in what AI does to unmonitored academic work. Serrano's midterm and final tested the same cohort on comparable material within a single semester. The only material variable was proctoring. A 50 percent decline in average score is not a marginal cheating signal.
The pattern matches survey data at peer institutions. A recent Princeton survey found that 29.9 percent of students admitted to using AI to cheat on at least one exam or assignment. A provost-led report at Brown documented that 56 percent of undergraduate respondents and 67 percent of graduate and medical students said they use generative AI tools daily or weekly, with large majorities also expressing concern about the effect on their own learning.
Serrano, who went blind from retinal dystrophy at 17 and went on to Harvard and later a career decorated with a 2025 award from the King of Spain, has refused to let the episode fade quietly. He has taken the story to El País and Inside Higher Ed and argued that Brown's administrative response has been muted.
The counterweight worth naming: take-home exams with unlimited time were always a porous format, and the ECON 1170 numbers reflect an extreme case in which a professor deliberately made the take-home harder to compensate. Some fraction of the score gap reflects time pressure and exam conditions rather than cheating alone. But the composition of who dropped — 22 of 27 departures being perfect midterm scorers — is difficult to explain on any other grounds.
For AI companies, the ECON 1170 case is the kind of data point that regulators and university administrators will cite for years. OpenAI, Anthropic, and Google all market education-tuned versions of their assistants, and each has published guidance on responsible academic use. Those guidelines have limited force against a student optimizing for a grade under deadline. Expect universities to accelerate the shift back toward proctored, in-person, and oral assessments — and expect the AI education market to bifurcate into tools that help students learn (tutoring, spaced repetition, feedback) and tools that help students submit (drafting, solving, summarizing). The second category is the one Serrano's numbers just put on trial.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
