Professor Roberto Serrano of Brown University allowed take-home exams in his difficult economics course ECON 1170 in spring 2026. The course attracted 86 students instead of the usual handful, and the midterm average reached 96 out of 100. Suspecting widespread AI cheating, Serrano switched the final to an in-person format. The average dropped from 96 to 48. Twenty-two of 27 students who withdrew or skipped the final had previously scored a perfect 100 on the midterm.
Key takeaways
- Take-home midterm average: 96/100 — historically ranges between 65-80 in this course
- After switching to an in-person final, the average dropped to 48 — a 50% fall
- 18 students withdrew, 9 did not show up for the final — 22 of 27 had scored a perfect 100 on the midterm
- Brown University report: 56% of undergrads use GenAI daily or weekly, most worried about impact on their cognitive skills
- Princeton survey: 29.9% of students admitted to cheating with AI on at least one exam or assignment
Record scores that did not add up
Roberto Serrano has taught ECON 1170 for years. He never had more than 30 students — usually a small group of strong ones. In spring 2026, 86 students enrolled, which was already unusual. Serrano believes the draw was the format: take-home exams, perceived as offering more freedom and time.
The March 5, 2026 midterm results were unprecedented. The average was 96 out of 100 — historically this course averages between 65 and 80. Forty students scored a perfect 100. When Serrano and his graduate students ran the exam questions through ChatGPT, the answers matched what students had submitted — with the same convoluted style that is a recognizable fingerprint of generative AI tools.
The verification test
Serrano sent his class a message: I am not declaring the midterm void yet. I am giving the class a chance to prove me wrong. If the final exam score distribution roughly mirrors the midterm, I will count it. Otherwise — which is what I expect — I will declare the midterm void and reweight the final accordingly.
Eighteen students immediately dropped the course. Nine others did not show up for the final. Twenty-seven total — 22 of whom had scored a perfect 100 on the midterm. Those who sat the in-person final averaged 48 out of 100. A drop of half.
A phenomenon wider than one course
The Brown case is not isolated. A survey at Princeton found 29.9% of students admitted to cheating with AI on at least one exam or assignment. A 2026 Brown University provost?Provost: The provost — a senior academic officer of a university responsible for teaching and research affairs. report shows that 56% of undergrads and 67% of graduate and medical students use generative tools daily or weekly.
The paradox is clear: the same people who use AI heavily also express deep concern about its impact on their cognitive abilities. The Brown report explicitly cites student concerns about the negative impact of GenAI on their learning. AI cheating appears to stem not from lack of skill, but from incentive structure: if everyone does it and detection is hard, the individual calculus is obvious.
The university response and the professor's stance
Serrano is frustrated by what he describes as a cautious response from Brown's administration. He has been blind since age 17 due to retinal dystrophy — and describes his approach to learning as a response to constraints, not capitulation. His willingness to go public comes from the conviction that silence is worse.
As quoted by Inside Higher Ed: We cannot afford to have a society in which a significant fraction of our best young minds think that cheating is okay. That leads to a declining society, to a failed society. We cannot choose to become idiots.
Why this matters
The Brown University scandal is not the story of an exception — it is a probe into the scale of a systemic problem. AI has made cheating nearly undetectable, cheap, and resistant to standard control methods. The only verification that worked for Serrano was returning to an exam room and a sheet of paper.
For universities, this is an institutional dilemma: how to maintain the value of a degree when a tool capable of solving any homework assignment is in every student's pocket? The answer through bans does not work — AI is ubiquitous. The answer through redesigning assessment — toward oral exams, live projects, context-dependent work — is expensive and resource-intensive.
A second risk is emerging in the broader labor market: degree inflation. Employers are already forced to verify candidate competencies differently than before, because academic grades have lost some of their signaling value. This is only the first chapter in a longer story about how generative AI is changing the relationship between educational institutions and knowledge.
What's next
- Brown University is working on AI-in-education recommendations — the 2026 provost report is a precursor to institutional policy, details not yet announced
- Roberto Serrano has pledged to continue speaking out — in his view only public pressure can push university administrations to act
- The Brown case will serve as a reference point in debates about academic assessment reform — more universities are considering switching to in-person or oral exams as the default
Sources
- Ars Technica — Suspecting AI cheating, Ivy League prof ordered an in-person final; scores fell 50%
- Inside Higher Ed — Brown professor suspects most of his class used AI to cheat
- Brown University — Generative AI in Teaching and Learning (provost report 2026)
