UNAM forces 58,000 to retake entrance exam after AI-proctored test yields 5x spike in top scores

Mexico’s largest university is in full damage-control mode after its first experiment with fully remote, AI-proctored entrance exams imploded spectacularly. Nearly 160,000 applicants took the test using a lockdown browser and webcam monitoring software. The results that came back were statistically absurd.

The numbers don’t lie. Between 2021 and 2025, a consistent 3.5% of test takers scored 100 or above on the 120-question exam. This year? That figure rocketed to 16.3%. The distortion at the very top is even more damning: the share of applicants hitting 110 or more jumped from a historical average of 0.9% to 5.5%. That’s not a cohort of geniuses suddenly emerging — it’s a systemic integrity failure that makes the entire admissions cycle unusable.

UNAM convened an expert commission with a characteristically bureaucratic name to investigate. Their recommendation landed like a hammer: a mandatory in-person “control exam” for roughly 58,000 people. This isn’t just for those who already secured a spot through the tainted test. The commission is casting a wider net, applying the retest to anyone who would have been admitted based on minimum score thresholds in their program dating back to 2021. The university rector apologized to honest applicants now forced to study and sit through another high-stakes exam, but insisted the drastic measure is necessary to “guarantee equity in access.”

The looming August 10 class start date makes this a logistical nightmare. UNAM must now design, schedule, and administer a mass in-person test — and process results — in a matter of weeks. This disaster is a flashing red warning for the broader edtech industry. AI proctoring has been sold as scalable and cheat-proof, but UNAM’s experience suggests that when the stakes are this high, remote monitoring can be catastrophically porous. The gap between the software’s promise and the statistical reality of the results is the strongest argument yet that no algorithm, however well-trained, can replicate the security of a human proctor in a physical room when a university seat is on the line. Classes will almost certainly be delayed.