AI Proctoring Botched 58,000 Entrance Exams at Mexico's Top University

The National Autonomous University of Mexico (UNAM) annulled 3,174 exams and halted all admissions after AI-powered proctoring software failed to prevent widespread cheating on its first fully remote entrance exam, forcing 58,000 students to retake the test. Nearly 160,000 applicants took the exam, and scores of 100 or above jumped from 3.5% in 2021–2025 to 16% this year, with nearly half of the 158,712 exams potentially involving cheating, according to a technical expert. UNAM rector Leonardo Lomelí apologized, and Mexico's president Claudia Sheinbaum suggested the attorney general may investigate.

Mexico's largest university handed its most competitive entrance exam to AI surveillance software—and the machines lost. Now 58,000 students have to retake the test because algorithmic proctoring couldn't stop a cheating free-for-all. This is what happens when institutions outsource human judgment to algorithms. The National Autonomous University of Mexico UNAM ran its first-ever fully remote entrance exam this year, relying on two pieces of software: LockDown, a browser by Respondus that locks down a test-taker's computer, and an AI-powered proctoring program by Territorium that used voice and facial recognition via webcam to monitor candidates. Nearly 160,000 applicants sat for the exam, competing for just 21,962 spots. The software failed on a grand scale. The numbers are damning. Between 2021 and 2025, only 3.5% of applicants scored 100 or above on the 120-point exam. This year, that figure shot to 16%. The share of students scoring 110 or above jumped sixfold, from 0.9% to 5.5%, according to the Guardian. In the medicine program—among the most competitive in the region—nearly 30% of candidates hit 100 or above, compared to just 9% last year. A technical expert who reviewed the results told the New York Times that nearly half of the 158,712 exams could have involved some degree of cheating. The AI couldn't catch it. Test questions were leaked and sold online while the exam was still live in May and June. Cheating tips circulated widely before the test opened. Gizmodo reported that one expert estimated at least half the participants cheated. Students protested at UNAM's Mexico City campus, chanting, "We demand honesty, not artificial intelligence." The university annulled 3,174 exams—about 2%—after detecting irregularities including identity impersonation, external assistance, and use of AI. But the real damage was far broader. UNAM's rector, Leonardo Lomelí, conceded the disaster: "The Unam, the home of science, humanities and understanding, knows how to recognize mistakes and self-correct." He apologized to students who took the exam fairly but will now have to sit for a control exam in person. All admissions have been halted—an unprecedented measure in the university's 115-year modern history. Mexico's president, Claudia Sheinbaum, suggested the attorney general may investigate. UNAM has filed a criminal complaint over alleged fraudulent services offered to applicants and their families. The university's stated motive was equitable access: students previously had to travel hundreds of miles to Mexico City to take the exam. A fair concern. But the remote AI solution created a worse problem—massive fraud that punished honest test-takers and invalidated the entire process. The same people selling AI surveillance as objective and efficient just torched the futures of tens of thousands of young people who now have to sit the exam again—this time, in person, the old-fashioned way. The question isn't whether technology has a role in education. It's who pays when the technocrats get it wrong, and why institutions keep trusting machines before they've earned it.