Brown University Professor Exposes Mass AI Cheating After Exam Scores Plunge From 96% to 48%

A Take-Home Exam That Revealed Too Much

When Roberto Serrano decided to offer take-home exams for his advanced economics course at Brown University in spring 2026, he made the choice for compassionate reasons. A gunman had attacked the campus in December 2025, killing two people, including someone who had recently introduced herself to Serrano 1

1

. Many students felt anxious about sitting in crowded classrooms, so the professor adapted his ECON 1170 course—Welfare Economics and Social Choice Theory—to allow both a take-home midterm exam and final 3

3

.

What followed became one of the starkest illustrations of AI cheating in higher education. The course, which typically attracts fewer than 30 students and sometimes as few as eight, suddenly enrolled 86 students 1

1

. The midterm results were extraordinary: an average score of 96 out of 100, with 40 students achieving perfect scores 2

2

. These unusually high midterm scores stood in sharp contrast to the historical average for the course, which had ranged between 65 and 80 percent 4

4

.

The Signs That Something Was Wrong

Serrano, a blind economics professor who has taught at Brown University for decades after earning his doctorate from Harvard, immediately sensed something amiss. He had deliberately made the exam harder than usual, reasoning that a take-home format with unlimited time was "an opportunity to challenge the class a little bit more" 1

1

. Yet many answers, even when correct, exhibited a "very convoluted style" that felt off 3

3

.

When Serrano and his teaching assistants ran the exam questions through ChatGPT, the generative AI tool produced solutions remarkably similar to what numerous students had submitted. The large language model used the same odd, roundabout approach to solving problems rather than straightforward direct proofs 4

4

. Rather than relying on AI-proofing tools, which often produce false positives and have prompted lawsuits from students wrongly accused, Serrano chose a more direct method to test his suspicions 2

2

.

Setting a Trap to Prove Mass Cheating

Serrano informed his class that the final would be an in-person final exam, and he would compare the two distributions. If the scores matched, he would count the midterm. If not, he would void it and reweight the final accordingly 1

1

. The response was telling. Eighteen students suddenly dropped the course, while nine others didn't attend the final. Of those 27 students, 22 had scored a perfect 100 on the midterm 3

3

.

Among the 59 students who took the in-person test, the average score plunged to 48.6 percent—a figure that had never fallen below 65 percent in the course's history 2

2

. Three students scored zero, and charting the results revealed that most students fell more than 30 points behind their midterm scores. Only two students appeared to have taken both tests without AI-assisted cheating: one scored 95.5 percent on the midterm and 95 percent on the final, while another posted 55 percent and 59 percent respectively 2

2

. By Serrano's count, at least 50 students cheated on the take-home midterm exam, and he considers the evidence overwhelming 3

3

.

A Professor Who Won't Back Down

Serrano, who went blind at age 17 from retinal dystrophy while growing up in Madrid, has never been inclined to coddle elite students. After a brief crisis following his diagnosis, he learned Braille and earned admission to Harvard. "We economists understand reality as a set of people responding to optimization problems with restrictions. I view my disease simply as one more restriction that I have to deal with, and I optimize based on that," he told El País 1

1

. His father would help him reconstruct classroom notes so Serrano could transcribe them into Braille, teaching him that "learning and succeeding doesn't come without effort" 4

4

.

This background shapes his uncompromising stance on the AI cheating scandal. After submitting his data to Brown's Standing Committee on the Academic Code and receiving no response, Serrano went public, telling his story to El País and Inside Higher Ed 2

2

. The academic committee later asked for individual complaints against each student and copies of their exams, but Serrano suspected they would simply run them through AI detection software—the very approach he had avoided 2

2

. "We cannot afford to have a society in which a significant fraction of our best young minds think that cheating is okay," Serrano told Inside Higher Ed. "That leads to a declining society, to a failed society. We cannot choose to become idiots" 3

3

.

The Broader Crisis in Academic Integrity

The AI cheating scandal at Brown University reflects a widespread crisis. A recent survey found that 29.9 percent of Princeton students admitted to cheating with AI on at least one exam or assignment 1

1

. Brown's own provost-led report on generative AI in teaching and learning found that 56 percent of undergraduate respondents and 67 percent of graduate and medical students reported using generative AI tools daily or weekly 1

1

. Yet large majorities also expressed concerns about cognitive impacts on their learning and feared negative consequences for their "cognitive capacity" 3

3

.

The widespread student use of generative AI has forced universities to scramble for solutions. New devices complicate enforcement: innocent-looking calculators with cameras and AI promise "Snap. Solve. Done," while eyeglasses embedded with cameras can provide answers without professors detecting anything 4

4

. Laurent Lessard, an associate professor at Northeastern University who served on an AI assessment task force, noted that AI could complete a semester-long college course in about two hours 4

4

.

Universities Adapt to the New Reality

Institutions across the Ivy League and beyond are resurrecting old-school practices. Stanford and Princeton universities began allowing—and in Princeton's case requiring—proctored exams despite their honor codes 4

4

. The University of Chicago's law school is piloting a device ban in first-year core classrooms, while some professors have revived blue books and oral exams 4

4

. Others have stopped assigning or grading homework, emphasizing the learning process over final products.

Brown released its committee report the same week Serrano's story went public, recommending changes including examining academic codes to ensure academic integrity in the age of AI remains central 4

4

. Brian Clark, a university spokesman, stated that "Brown treats every allegation of academic integrity with the utmost seriousness" 4

4

. Yet experts suggest the problem will persist as long as no one is paid specifically to ensure student integrity, and that analog, in-person testing may be the only reliable deterrent 2

2

.

What This Means for the Future

Serrano's chart showing the dramatic score collapse from 96 to 48 percent turns abstract concerns about the erosion of academic integrity into hard data. As one analysis noted, "Take the AI away, and half the apparent knowledge goes with it" 3

3

. The question facing higher education is whether students competing for jobs and graduate school admissions can resist tools that promise better grades with minimal effort. A February Pew Research Center report found that 59 percent of teens believe using AI to cheat is regular at their school 4

4

, suggesting this pattern may intensify.

The stakes extend beyond individual courses. If elite students substitute ChatGPT for genuine learning, they risk what Serrano frames as societal failure—a generation that appears educated on paper but lacks the cognitive capacity to solve complex problems . Universities must now decide whether to adapt assessment methods to an AI-saturated world or fight to preserve spaces where human learning still matters. The answer will shape not just academic standards, but the capabilities of the workforce and leadership these institutions produce.