Brown AI Cheating: 50 Students, Scores Drop from 96% to 48%

For EdTech professionals, the largest AI cheating case in Ivy League history is a wake-up call. When a Brown University professor moved exams online after a campus tragedy, at least 50 students exploited the format to use AI tools, achieving an improbable 96% average on a difficult midterm—only to plummet to 48% on an in-person final. The incident underscores the critical demand for innovative proctoring, AI-resistant assessment, and integrity platforms that can keep pace with generative AI.

In what may be the largest artificial intelligence cheating scandal in Ivy League history, Brown University economics professor Roberto Serrano has publicly condemned his institution's inadequate response after uncovering that at least 50 of the 86 students in his spring 2026 advanced economics course used AI tools, likely ChatGPT, to fraudulently complete a take-home midterm exam. The incident has exposed significant gaps in academic integrity enforcement at elite institutions, the limitations of current AI detection methods, and the profound challenges that generative AI poses to traditional assessment models. The events trace back to a tragic campus shooting in December 2025, which prompted Serrano to move his traditionally in-person exams to an online, take-home format to reduce student anxiety. That decision triggered an enrollment surge from the typical 30 students to 86, as students apparently perceived the new format as an easier path. The midterm, which Serrano described as more difficult than in previous years, saw the class average soar to an astonishing 96 percent—far above the historical range of 65 to 80 percent—and 40 students achieved perfect scores. Suspicious of the anomalously high results, Serrano and his teaching assistants fed the exam questions into ChatGPT and found that many student answers closely mirrored the AI's outputs, exhibiting similar logic structures and unusual phrasing. The professor concluded that at least 50 students had engaged in academic dishonesty, making it the most significant known case of its kind at Brown and a landmark event for higher education.

“

When a Brown University professor moved exams online after a campus tragedy, at least 50 students exploited the format to use AI tools, achieving an improbable 96% average on a difficult midterm—only to plummet to 48% on an in-person final.

In response, Serrano reverted the final exam to an in-person, proctored format. He warned students that if their final scores did not match their midterm performance, he would void the midterm results. The consequences were immediate: 18 students dropped the course before the final, and another 9 enrolled students did not take the exam. Among those who did sit for the final, the average score plummeted to 48 out of 100, and only a handful performed at a level consistent with their earlier inflated results. This stark reversal provided compelling evidence that widespread cheating had occurred. Despite the overwhelming data, Serrano's initial submission to Brown's Standing Committee on the Academic Code was met with silence. Only after he publicized the matter did the committee request that he file individual complaints for each suspected cheater—a process Serrano criticized as impractical, particularly given the well-documented unreliability of AI detection software and the difficulty of proving intent in a take-home setting.

The incident highlights a systemic crisis in academic integrity. The rapid advancement and accessibility of generative AI tools have far outpaced institutional policies and the technical capabilities of detection software. Traditional plagiarism checkers are ill-equipped to identify AI-generated text, and emerging AI detection tools suffer from high false-positive rates, making them legally and ethically problematic for imposing sanctions. Brown's procedural demand for individualized complaints underscores a broader institutional reluctance to confront AI-assisted cheating collectively, effectively shifting the burden onto faculty and creating a strong disincentive to report cases. This case also demonstrates the powerful incentive structures that online and take-home assessments create: students quickly recognized that AI could produce high-quality answers with minimal effort, turning what was intended as a compassionate accommodation into an opportunity for mass academic fraud.