AI Fraud at Brown University: The Collapse of Take-Home Exams

Mass AI Fraud at Brown University

Professor Roberto Serrano of Brown University has uncovered evidence that at least 50 students used artificial intelligence to cheat on a midterm exam in ECON 1170, an advanced undergraduate course in mathematical economics. This incident represents one of the largest known academic integrity scandals in the Ivy League, signaling a critical failure of traditional take-home assessment models in the era of Large Language Models (LLMs).

Professor Serrano, the Harrison S. Kravis University Professor of Economics, detected the fraud after a midterm exam administered on March 5 resulted in an extraordinary average score of 96 out of 100, with 40 students achieving a perfect score. The fraud was confirmed through two primary methods:

Pattern Analysis: Grading assistants noted unusual passages in student answers that mirrored results generated by ChatGPT.
Performance Divergence: When Serrano shifted the final exam (worth 50% of the grade) to an in-person, proctored format, the average score plummeted to 48 out of 100. Furthermore, 22 of the 27 students who failed to show up for the final exam had scored a perfect 100 on the AI-suspected midterm.

The Institutional Response and Academic Tension

The administration at Brown University has been criticized by Professor Serrano for its perceived indifference toward the scale of the fraud. Serrano reports that the university president remained silent and the dean offered no comment until the case was brought before the Academic Code Committee, which eventually described the event as a "wake-up call."

Serrano argues that the university's reluctance to act may be tied to the influence of wealthy donors whose children attend the institution, suggesting that students often receive the benefit of the doubt regardless of evidence. He maintains that without a public admission of the problem and a broad debate on academic integrity, the prestige and utility of higher education are at risk.

The Death of the Take-Home Exam

The rise of AI is forcing elite universities to abandon long-standing academic traditions in favor of strictly proctored, in-person testing. The "take-home, closed-book" exam—a tradition at many Ivy League schools designed to allow for more complex, time-intensive problems—has become functionally impossible to secure.

This shift is visible across other institutions as well. Princeton University recently ended a 133-year-old tradition of unproctored exams based on an Honor Code, returning to a system where professors must physically proctor tests to prevent AI-assisted deception.

Expert Perspectives on AI and Pedagogy

Discussion among academics and students suggests that the problem is not merely the technology, but the misalignment of incentives in high-pressure academic environments.

The Incentive Gap

Many observers argue that students in competitive programs are forced into a "prisoner's dilemma" where cheating becomes the game-theoretic optimal choice. If a course is graded on a curve and a significant portion of the class uses AI to achieve perfect scores, honest students are effectively penalized for their integrity.

"When you're a student in a competitive program at a top university, graded on a curve, and you know your fellow classmates are cheating with AI, you have little choice but to do the same." — Hacker News contributor @pants2

Proposed Solutions for Assessment

Educators are proposing several shifts in how student competency is measured to mitigate AI fraud:

Adversarial Course Design: Designing curricula so that the path to a high grade requires meeting learning objectives regardless of AI use. This includes using 1-on-1 interviews to verify that students understand the code or proofs they submit.
Process-Based Assessment: Moving away from grading the final output and instead assessing the process of creation, or using structured video reports where students must defend their design decisions.
Hybrid Testing: A mix of handwritten "blue book" exams, oral examinations, and AI-assisted essays to ensure a comprehensive evaluation of skill.
Randomized Validation: Using ungraded homework where a single problem is randomly selected for a proctored in-class quiz, forcing students to engage with all the material.

Conclusion

Professor Serrano has fundamentally altered his teaching approach for the coming year: weekly exercises will no longer count toward the final grade, and take-home exams have been permanently abolished. The Brown University incident serves as a case study for the broader necessity of decoupling academic credentials from outputs that can be generated by AI.