Professor noticed class averages jump from 65% to 96% and tricked ChatGPT into revealing class-wide cheating. His next move made 18 students drop out
Roberto Serrano changed one rule in his classroom and found that his whole class was cheating
When the spring semester grades rolled in for Brown University's "Welfare Economics and Social Choice Theory" course, the results looked historic: the class averaged a staggering 96% on its midterm. Dozens of undergraduates secured flawless, perfect scores. But for Roberto Serrano, a seasoned economics professor who had spent nearly two decades teaching the complex curriculum, the flawless data didn't signal a sudden influx of geniuses. He recognized that the students were using Artificial Intelligence to assist with their work and decided to demonstrate this point, as reported by Inside Higher Ed on July 8, 2026.
A Brown professor gave his students a take-home midterm exam. After suspecting many cheated using AI, he made the final in-person. The orange dots are the midterm scores and the gray dots are the final scores. Looks like all but 3 cheated on the midterm. pic.twitter.com/hekcGsz76h
— Paul Graham (@paulg) July 9, 2026
A leniency born out of tragedy
Following a tragic shooting near the Brown University campus in December 2025, Professor Serrano decided to adjust his pedagogical approach for the upcoming spring term. Seeking to alleviate heightened student anxiety and trauma, he broke a strict 18-year-old rule against take-home midterms for his notoriously difficult course. This shift coincided with a massive, unprecedented spike in enrollment. His classes, which typically averaged around 30 students, ballooned to a staggering 86.
Suspiciously perfect performance
When Serrano graded the take-home midterms, the results far exceeded his usual results. The class achieved a 96% average. "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," he revealed. This was enough for him to clock that something fishy was going on. So, he teamed up with his graders and ran his own exam prompts through ChatGPT and watched AI spit out the same answers found in the majority of the students' submissions.
The final exam pivot
After explaining the results to the Dean, Serrano obtained permission to administer the final exam in person. Obviously, this triggered a wave of panic amongst the students, with 18 of them straight up dropping out of the course. Meanwhile, another nine skipped the test entirely to avoid being caught, and three students who did show up scored zero after failing to answer even one question. Nonetheless, for those who did complete the test without any digital assistance, the scores plummeted to a historic 48.6% class average.
Serrano noted that this drastic statistical collapse provided "overwhelming evidence" that the midterm had been compromised by widespread AI use. At the same time, following this, he declared the midterm void and announced to his students that the final exam would be worth 80% of their final grade. Anyone who scored 40% or higher earned a passing grade, which previously was 50% or higher. Overall, about 19 students failed the class.
Data revealed widespread misuse
The crisis unfolding at Brown University reflects a sweeping national trend. A massive study by the Student Experience in the Research University Consortium and Cornell University reveals that 37% of undergraduates regularly use generative AI, with 9% admitting to using it to cheat in exams. This misconduct is highly concentrated amongst daily users, where the cheating rate climbs up to a staggering 26%. The data also shows that economics students are most likely to cross this ethical line, with a 17% cheating rate.
A 'meek' institutional response
Despite the glaring disparity and evidence of cheating, Serrano submitted his results in May in front of the University's committee handling the situation, because the university's honor code requires a high burden of proof for individual cases, and pursuing dozens of separate academic dishonesty charges was nearly impossible. That's why later he criticized the administration's stance as "meek," warning that the lack of decisive systemic action against generative AI cheating undermines the value of an Ivy League degree and leaves faculty members fighting a losing battle alone.
This is a disservice
While I agree that students "cheating" on tests in college is ultimately just doing themselves a disservice, I also think it's the responsibility of those who are selling them an education to figure out the best way to actually teach them.
— Johnathon (@Synnoia) July 10, 2026
I would conclude that 2/3 of the class probably cheated, in some fashion. Final exams are often comprehensive, and students often fare worse when exams cover a wide range of material.
— Dr. Alchemy (@Dr_Alchemy) July 10, 2026
Shortly after Serrano went public with his findings, the conversation spilled over to X, where a man named Paul Graham (@paulg) posted the chart of the scores, and people had mixed reactions. @xorxorg wrote, "If the midterm was take-home, these scores reflect textbook research and googling, not necessarily AI." In contrast, @gseimon commented, "Degrees are even more worthless now that AI is doing an out thinking and students are getting degrees for cheating."
More on Scoop Upworthy
High school student takes the high road after learning valedictorian 'always cheats'
Share on Facebook
