Mexico's largest university has quietly produced one of the biggest automated-proctoring failures on record. The National Autonomous University of Mexico (UNAM), the country's top institution and one of the most prestigious in Latin America, has ruled that nearly 58,000 applicants must retake its entrance exam. This time there will be no testing from home: the redo will be in person, in an official exam room, watched by a human proctor. The reason is uncomfortable — officials concluded that the remote stage, which leaned on AI proctoring, produced results they simply cannot trust amid signs of large-scale cheating.
The decision caps an experiment that ran from May into early June. For the first time in the university's history, roughly 160,000 candidates sat the exam entirely from home, using a “lockdown” browser, facial recognition and AI-powered webcam proctoring software. The pitch was simple: replicate, at scale, the invigilation that once required classrooms, supervisors and attendance sheets. The problem is that the technology, built to catch wandering eyes and outside sources, failed to stop exactly what it feared most.
When results were published on July 17, the score curves looked like no previous year. Between 2021 and 2025, only 3.5% of test-takers scored 100 or higher on the 120-question exam. This time, the share of top scores rose by roughly five times, according to analyst estimates. Scores were so high, so concentrated and so uniform that the leading explanation became AI-assisted cheating — not by humans in the room, but by generative assistants able to reason through almost any question in seconds.
The episode exposes a weakness the proctoring industry would rather downplay: automated supervision watches the body, not the mind. It flags movements, sounds and whether faces stay in frame, but it struggles to tell a nervous student from one quietly consulting a chatbot out of view. On standardized multiple-choice tests, AI becomes a fraud multiplier: anyone who cannot be meaningfully observed can, in theory, outsource the answer to every single question. It is an asymmetric arms race, with examiners policing the surface while candidates exploit the gap.
The reliance on such tools has grown sharply since the pandemic, when campuses went online almost overnight, and it never really receded. UNAM's gamble was among the boldest leaps yet — trusting off-the-shelf surveillance to guard a make-or-break admission process that decides the future of tens of thousands of teenagers. That bet, in the span of a single season, turned into a public-relations problem and a logistical one: organizing supervised, in-person retakes for tens of thousands of people is not a weekend project.
The human toll is heavy. Roughly 58,000 people must prepare again, wait for a new date and walk into a room full of supervisors after months of study. For many, this is about fairness and opportunity; for the university, the rector justified the control exam as “necessary to give certainty and guarantee equity in access.” The cost in anxiety, travel and lost preparation lands hardest on those who needed the chance most.
That leaves the open question: is the answer to abandon the AI, or to abandon the model itself? Perhaps the lesson is not that the technology failed, but that it should never have been the only line of defense for academic integrity on a high-stakes exam. If a supervision that supervises nothing produces results no one believes, then who validates the validator?
Sources: Ars Technica, Le Monde, Crypto Briefing
✓ Independent sources cross-checked and verified before publishing
