An economics professor at Brown University has uncovered what he calls overwhelming evidence of mass cheating using artificial intelligence in his advanced undergraduate course. Roberto Serrano, who teaches ECON 1170, watched his take-home midterm average soar to an unprecedented 96 out of 100. When he switched the final exam to an in-person format, the average collapsed to 48. The stark disparity has prompted Serrano to go public with his findings, sparking renewed debate about the role of AI in academic integrity.
Serrano’s decision to offer take-home exams this term came from a humane impulse. After a gunman killed two students on campus last December, many students expressed anxiety about sitting exams in crowded rooms. To ease their distress, Serrano allowed take-home midterm and final papers for the first time in his decades-long career. The irony was bitter: the moment he relaxed the rules, a significant portion of the class appears to have turned to AI for answers.
The telling statistics
ECON 1170 is an advanced economics course that typically draws a small, dedicated cohort. Serrano had never taught more than 30 students; one term he had only eight. This semester, however, 86 students enrolled. The sudden spike, Serrano believes, was driven largely by the new take-home format, which made the course more attractive to students seeking an easy path to high grades.
The midterm results were, in Serrano’s own word, extraordinary. The class average hit 96, with 40 students achieving a perfect score of 100. Historically, the course average has ranged between 65 and 80. Moreover, this year’s midterm was deliberately more difficult than usual, designed to push students further given the unlimited time they had at home. The disconnect between difficulty and scores immediately raised red flags.
Beyond the numbers, the quality of the answers felt off. Many correct responses exhibited what Serrano described as a “very convoluted style.” When he and his graders fed the exam questions into ChatGPT, the AI produced strikingly similar answers. Patterns of phrasing, logical jumps, and even specific errors matched what the AI generated.
The decisive test
Determined to prove his suspicions, Serrano set a trap. He announced to the class that the final exam would be held in person under traditional proctoring. He further stated that he would compare the score distributions of the midterm and final. If they matched, he would retain the midterm grades. If not, he would void the midterm entirely and reweight the final to determine the course grade.
The results were damning. Eighteen students immediately dropped the course. An additional nine skipped the final exam altogether. Of those 27 students, 22 had scored a perfect 100 on the midterm. Among the students who did take the final, the average plummeted from 96 to 48—a drop of 50 percent. By Serrano’s estimation, at least 50 of the 86 students cheated on the midterm using AI. He called the evidence overwhelming.
A broader crisis in academia
Brown University is far from alone in confronting AI-enabled cheating. A recent survey of Princeton students revealed that nearly 30 percent admitted to cheating on at least one exam or assignment, with the vast majority using AI tools like ChatGPT. Across the United States, colleges and universities have spent the past two years scrambling to deploy detection software, revise honor codes, and rethink assessment methods altogether.
The challenge is particularly acute in courses that rely on take-home essays or problem sets. Generative AI can produce coherent, well-structured text and solve complex mathematical problems, making it nearly impossible for instructors to distinguish student work from machine output without rigorous controls like in-person exams or oral defenses. Some institutions have returned fully to paper-based, proctored exams; others are experimenting with AI-proof assignments that require personal reflection, real-time collaboration, or integration of multimedia sources.
Students themselves report feeling conflicted. Brown’s own provost-led task force found that the majority of undergraduates use generative AI weekly or even daily, both for academic and personal tasks. Yet large majorities also expressed concern about the technology’s impact on their learning. Many worry that overreliance on AI is eroding their cognitive capacity—their ability to reason, write critically, solve problems independently, and retain knowledge. This phenomenon, sometimes called “cognitive offloading,” is a growing focus of educational research.
The stakes for society
Serrano frames the issue in stark terms. “We cannot afford to have a society in which a significant fraction of our best young minds think that cheating is okay,” he told reporters. “That leads to a declining society, to a failed society. We cannot choose to become idiots.”
His experiment is small—one class, one term—but it crystallizes a fear that many educators hold. When the AI crutch is removed, half of the apparent knowledge evaporates. That figure forces universities to confront uncomfortable questions. Are they awarding credentials that no longer reflect genuine competence? Are they preparing students for a world where AI is ubiquitous, or allowing them to bypass the very struggle that builds expertise?
Some argue that the solution is not to ban AI but to integrate it responsibly into curricula. Professors at institutions like Stanford and MIT are redesigning assignments to require students to critique AI-generated outputs, combine AI suggestions with original analysis, or use AI as a brainstorming partner while demonstrating mastery through in-person presentations. Others advocate for a return to oral exams, blue-book essays, and proctored assessments that leave no room for technological shortcuts.
Yet the pace of change is rapid. New models emerge every few months, each more capable than the last. Detection tools lag behind, and students quickly learn which prompts evade algorithmic scrutiny. The cat-and-mouse game between cheaters and enforcers shows no signs of resolution.
For Serrano, the immediate lesson is clear: take-home exams are no longer feasible in an era of accessible AI. But the deeper issue is one of values. “If we allow the best and brightest to believe that cutting corners is acceptable, we will produce a generation that can’t think its way out of a paper bag,” he warned. “That’s not just an academic problem. It’s a societal catastrophe.”
The numbers from Brown offer a stark data point. Take the AI away, and half the apparent knowledge disappears. That is the figure universities now have to sit with—and act on.