Brown Professor Alleges AI-Assisted Mass Cheating in Exam

Brown Daily Herald reported in April that a recent exam in Roberto Serrano's economics class had a median of 98%, with 40 of 86 students scoring 100. Inside Higher Ed, Ars Technica, Fortune, and El Pais later reported Serrano's allegation that many students used AI on the take-home midterm and that performance fell sharply after he moved the final exam in person.

This is an evaluation problem as much as a conduct problem. If a task can be solved by prompting a general model and lightly editing the output, then the assessment may be measuring access, prompt skill, and risk tolerance rather than independent reasoning.

Academic-integrity tools should avoid claiming certainty from a single signal. Stronger systems combine supervised settings, oral defense, version history, process artifacts, constrained tool policies, and human review. Similar design logic applies to workplace certifications, coding screens, and any high-stakes evaluation where LLM access is hard to exclude.

The institutional response matters because universities are likely to standardize new assessment policies before detection technology becomes reliable enough to stand alone.