UNAM Entrance Exam AI Proctoring Fails, Suspected Mass Cheating

The National Autonomous University of Mexico (UNAM) implemented an artificial intelligence proctoring system for its highly competitive entrance exam for the first time this year, a move that has led to widespread suspicion of mass cheating among approximately 160,000 applicants. UNAM, which admits around 50,000 students annually, reserves more than half of its enrollment slots for students from its 14 affiliated prep high schools. The remaining 22,000 spots are filled by external applicants who must pass a rigorous entrance examination. Historically, this exam was administered in person with human proctors overseeing the test-takers. This year, the university transitioned the exam to an online format, aiming to enhance accessibility. The online system incorporated two primary software components to ensure academic integrity. The first was LockDown, a specialized browser designed to restrict users from accessing any other applications or websites on their computers during the examination period. The second component was an AI proctor, developed by a company named Territorium, which utilized applicant webcams to detect and flag any indications of cheating. However, evidence suggests the AI system may have been significantly ineffective in preventing dishonest practices. Prior to this year's exam, the proportion of participants scoring 110 or higher had consistently remained at a low 0.9% over the preceding five years. In stark contrast, this year witnessed a dramatic increase, with 5.5% of applicants achieving scores of 110 or above, a more than five-fold surge. This substantial deviation from historical performance immediately raised red flags, prompting both applicants and external observers to question the exam's fairness. According to a report by The New York Times, AI specialist and mathematician Raúl Rojas conducted an analysis of the exam results. Rojas estimated that as many as 75,000 test-takers, representing roughly half of the total applicant pool, may have engaged in cheating. Rojas posited that a common cheating method, the "cheat sheet," was likely employed by these students. Despite the AI proctor's supposed vigilance in monitoring for indicators of cheating, such as the use of visible headphones or deactivated microphones, the system apparently failed to detect numerous instances of academic dishonesty, leaving ample opportunities for test-takers to circumvent the rules.