From 2021 through 2025, just 0.9 percent of applicants managed a score of 110 or better on UNAM’s 120-question entrance exam. This year the figure was 5.5 percent.
A six-fold leap at the very top of the curve, achieved in one year — the same year the test was first administered remotely. Nobody at Mexico’s largest university required a statistician to see that something had broken.
The consequence: about 58,000 people now have to take the exam over again, seated in a room, watched by a human.
The numbers that broke the exam
Close to 160,000 applicants sat the UNAM entrance exam earlier this summer, across several weeks running from late May into early June. It marked the first fully remote administration of the test, built around a lockdown browser and AI webcam proctoring.
What came back bore no resemblance to the previous four years of results. Across 2021 to 2025, 3.5 percent of test takers reached 100 or more. This year, 16.3 percent got past that mark.
Score distributions do not shift that way. Not over 160,000 people, not within 12 months, and not unless the conditions themselves have changed.
What the software was supposed to stop
Respondus’ LockDown Browser was mandatory for the test. The company says that once an exam is underway inside the browser, students are “unable to print, copy, go to other web addresses, or access other applications on the computer. Web searches, instant messaging, minimizing the browser, and hundreds of other functions are also prevented. Once the exam is completed, students are permitted to exit the browser and their computer will return to its normal state.”
That is a genuine lockdown, and it is the category’s standard product. It also governs exactly one machine.
On top of it, UNAM added a proctoring layer from Territorium, which ran AI algorithms over each candidate’s webcam feed. The system looked for a stand-in taking the seat in front of the computer, for phones or earphones, and for applicants stepping out of frame altogether. A single human supervisor was assigned to every 150 applicants, taking system alerts and passing irregularities up to UNAM officials.
The workarounds were public before the test started
A New York Times report published last week observed that “cheating tips were circulating widely before the tests went live this spring. Online accounts advised students to position monitors outside the web-camera frame where they could access ChatGPT or other AI models. Others suggested hiding earphones under their hair, or hiring someone else to take the exam, off camera.”
Line those three tactics up against the two products and the hole is plain. A lockdown browser locks down the computer the exam is running on. It has no opinion about the second monitor sitting two feet to the left. The webcam AI hunts for earphones — so you bury the earphones under your hair.
The defenses assumed a candidate alone with a single machine. That is not what the candidates were.
How much of it got caught
Nearly 2 percent of all exams were canceled by UNAM over conduct issues it has not specified, so the system did flag people. Weigh that against the university’s own admission: “the probability that a significant number of applicants may have received help” on the test.
AI expert Raul Rojas told NPR that statistical modeling pointed to “that almost half of the students were cheating when taking the exam online.”
Two percent detected versus a modeled 50 percent. Whatever the true number turns out to be, that is not a detection system that needs retuning. It is a detection system aimed at the wrong thing.
And what actually happened is still unknown. Because the exam was multiple choice rather than essay-based, the familiar AI-cheating signatures — a finished answer materializing in a text box in one go — simply do not exist here. Cheat sheets, leaked questions and the rest of the analog repertoire all remain live possibilities.
Who pays for it
A commission of experts was appointed by UNAM to investigate, burdened with the title “la Comisión Técnica de Personas Expertas para la Revisión del Proceso de Selección de Ingreso a Licenciatura para el Ciclo Escolar 2026-2027/1.” Its recommendation is in: an in-person control exam.
This year’s cohort is not the end of it. The retake covers everyone who would have earned admission on the basis of minimum successful scores in their program of study going back to 2021. Roughly 58,000 people fall inside that scope, and their places at UNAM now depend on the new result.
According to Gaceta UNAM, the university’s official news outlet, the rector has apologized to the honest applicants, who must now revise for and sit an exam a second time despite having done nothing wrong. Even so, the rector said the control exam remains “necessary to give certainty and guarantee equity in access.”
The clock is the other problem
Specifics on the control exam are expected shortly. As it stands, classes are set to start on August 10, which leaves virtually no window to schedule, staff and administer an in-person exam for 58,000 people — unless the university moves the start date.
Here is the piece worth dwelling on for any institution currently shopping for remote proctoring. UNAM paid for two products, staffed a human supervisor for every 150 candidates, and still could not certify one single score. Its fallback was not a partial audit or a re-scoring. It was a room, a proctor, and 58,000 people starting over.