UNAM Orders 58,000 Entrance Exam Retakes After AI Proctoring Breakdown

Summary

What happened, Mexico’s National Autonomous University, or UNAM, plans to require roughly 58,000 applicants to take an in-person “control exam” after its first fully remote undergraduate entrance examination produced anomalously high scores and allegations of widespread cheating. Nearly 160,000 people completed the 120-question test from late May through early June 2026 using a locked-down browser, automated webcam monitoring, and limited human oversight. A university-appointed expert commission recommended the retest, which will determine who receives a place for the 2026-2027/1 academic cycle.

Verified results, The strongest evidence of irregularities is the sharp departure from historical performance. From 2021 through 2025, 3.5 percent of candidates scored at least 100 out of 120; in 2026, that share rose to 16.3 percent. At the very top, the proportion scoring 110 or more increased from 0.9 percent in the previous five years to 5.5 percent this year. The retest will cover both candidates who initially secured admission and anyone whose 2026 score met the minimum admission threshold their chosen program had required in any year since 2021.

Security context, Candidates used Respondus LockDown Browser, which is designed to prevent activities such as opening other applications, visiting websites, copying material, printing, or minimizing the test window. Territorium’s AI-based proctoring system monitored webcams for possible impersonation, phones, earphones, or candidates leaving the frame. Human supervisors were assigned at a ratio of one for every 150 applicants and received automated alerts about possible irregularities. UNAM canceled nearly 2 percent of exams for unspecified conduct violations, showing that the controls detected some incidents but not enough to explain or contain the broader statistical anomaly.

What remains unproven, Neither UNAM nor its commission has publicly established exactly how much cheating occurred or which methods were responsible. The multiple-choice format makes some common signs of generative-AI use difficult to identify. Reported possibilities include off-camera monitors displaying ChatGPT or other AI tools, concealed earphones, unauthorized assistance, impersonation, cheat sheets, and leaked questions. AI specialist Raul Rojas told NPR that statistical modeling suggested almost half of online test takers may have cheated, but that remains an expert estimate rather than a confirmed count of individual violations. UNAM has more cautiously acknowledged the probability that a significant number of applicants received help.

Why it matters, Admission to Mexico’s largest university is highly competitive, so inflated scores could displace honest candidates and undermine confidence in the selection process. The remedy also imposes costs on applicants who followed the rules: they must prepare again, travel to an official testing location, and prove themselves under new conditions. UNAM’s rector apologized to those candidates while arguing that an in-person control exam is needed to restore certainty and equitable access. More broadly, the episode illustrates the limits of automated remote proctoring when prohibited assistance can occur outside a webcam’s field of view.

What happens next, UNAM is expected to publish detailed arrangements for the control exam, with approximately 58,000 admissions now dependent on its results. The timetable is exceptionally tight because classes were scheduled to begin on August 10, only a week after the August 3 report. It remains uncertain when the retest will occur, whether the semester will be delayed, how candidates unable to attend promptly will be accommodated, and whether UNAM will disclose a fuller technical account of why its browser, AI monitoring, and human-supervision system failed to protect the exam’s integrity.