Nearly 160,000 applicants sat for the entrance exam at UNAM, Mexico’s largest university, in a fully remote format over several weeks from late May through early June 2026, using a lockdown browser and AI-powered webcam proctoring. After results showed a sharp, unprecedented spike in top scores, the university has ordered an in-person control exam that will affect roughly 58,000 people, including some who earned admission in earlier years.<\/p>\n
For the first time, UNAM administered its undergraduate admissions test entirely online. Applicants had to install the LockDown Browser from Respondus, which blocks printing, copying, navigation to other websites, and access to other applications once an exam begins. UNAM also layered in a proctoring system from Territorium that used AI to monitor each test-taker’s webcam for signs of impersonation, cell phone or earphone use, or the applicant leaving the frame. One human supervisor was assigned for every 150 applicants, and those supervisors received alerts from the system so they could flag irregularities.<\/p>\n
The safeguards did not hold. Between 2021 and 2025, 3.5 percent of test takers scored 100 or more on the 120-question exam. In 2026, 16.3 percent did. At the very top of the scale, the picture was even more lopsided: 0.9 percent of test takers scored 110 or more between 2021 and 2025, but 5.5 percent did this year. Across five years, the share of top scorers rose by roughly fivefold, and the share of the highest scorers rose by roughly sixfold.<\/p>\n
UNAM canceled nearly 2 percent of total exams for unspecified conduct issues, a small fraction of the apparent irregularities. The university has acknowledged that a significant number of applicants may have received help.<\/p>\n
The exam was multiple choice, so the typical indicators of AI cheating, such as complete answers being pasted into text boxes, were not available to investigators. A New York Times report last week noted that cheating tips circulated widely before the test went live. Online accounts advised students to position monitors outside the webcam frame so they could access ChatGPT or other AI models, hide earphones under their hair, or hire someone else to take the exam off camera. Cheat sheets and leaked questions, more traditional methods, were also possible.<\/p>\n
AI expert Raul Rojas, speaking to NPR, said statistical modeling suggested that almost half of the students were cheating when taking the exam online.<\/p>\n
UNAM appointed a commission of experts, formally named the Comisi\u00f3n T\u00e9cnica de Personas Expertas para la Revisi\u00f3n del Proceso de Selecci\u00f3n de Ingreso a Licenciatura para el Ciclo Escolar 2026-2027\/1, to investigate the anomaly. The commission’s recommendation was to administer a control exam in person, with human proctors, to restore confidence in the results.<\/p>\n
The control exam will apply not only to those who secured a spot at UNAM based on this year’s test but also to everyone who would have been admitted based on minimum successful scores in their program of study since 2021. About 58,000 people could be affected, and places at UNAM will now depend on the results of the new test. The university rector has apologized to applicants who did nothing wrong but will have to take the test again, saying the control exam is necessary to give certainty and guarantee equity in access. Details on the retake were still pending publication; classes were scheduled to begin on August 10, so the timeline was tight.<\/p>\n
Remote exam delivery adds a layer of risk that in-person testing does not, and the UNAM experience shows how those risks can scale. The proctoring tools were designed to detect specific behaviors: another person in front of the camera, a phone or earphone in the frame, the applicant walking away. They were not designed to detect a second monitor just outside the camera’s view, hidden earphones, or a hired proxy sitting in for the registered applicant. A webcam-based system can only see what the camera sees, and a determined test-taker can move cheating outside that field of view.<\/p>\n
The sheer volume of test takers, nearly 160,000, also strained the human side of the supervision model. One supervisor per 150 applicants left little capacity to follow up on every alert the AI system generated, especially when the alerts were only a starting point for a closer look.<\/p>\n
UNAM’s experience is a high-stakes test of remote oversight, but it is not the only setting where remote work has been studied. A 2024 survey of academic health sciences libraries in the United States, published in the Journal of the Medical Library Association and led by David Petersen at the University of Tennessee Graduate School of Medicine, with co-authors Matthew Covey of The Rockefeller University and Janet Crum of Fresno State, found that remote and hybrid work was widely used and generally well received. Among 383 staff respondents in 2023, 78 percent were allowed to work remotely, and large majorities said remote work positively affected work-life balance (75 percent), morale and job satisfaction (69 percent), likelihood of staying at their institution (64 percent), and productivity (58 percent). That study measured attitudes and self-reported outcomes in a library setting, not test integrity, so its findings do not speak directly to the cheating problem at UNAM, but it does suggest that remote arrangements can work well when the underlying task is well matched to the medium.<\/p>\n
The UNAM case is a counterexample: high-stakes, multiple-choice testing, with strong incentives to cheat, is a poor match for unsupervised remote delivery even with AI proctoring layered on top.<\/p>\n
Top scores on UNAM’s 2026 entrance exam rose dramatically compared with the previous five years, and the university concluded that a significant number of applicants may have received help. To restore confidence in the results, UNAM’s expert commission recommended an in-person control exam that will apply to recent admittees and to applicants who would have qualified since 2021.<\/p>\n
UNAM required the LockDown Browser from Respondus, which blocks printing, copying, and other applications during the test, and a webcam-based proctoring system from Territorium that used AI to flag signs of impersonation, phone or earphone use, or the applicant leaving the frame. One human supervisor was assigned for every 150 applicants to review alerts.<\/p>\n
Between 2021 and 2025, 3.5 percent of test takers scored 100 or more on the 120-question exam and 0.9 percent scored 110 or more. In 2026, 16.3 percent scored 100 or more and 5.5 percent scored 110 or more, a roughly fivefold jump in the top tier and a roughly sixfold jump at the highest tier.<\/p>\n