UNAM Orders 58,000 Applicants to Retake Entrance Exam
Mexico's National Autonomous University (UNAM) ordered about 58,000 applicants to retake its undergraduate entrance exam in person after suspected widespread cheating during its first remotely administered test. The online exam used LockDown Browser and AI-assisted webcam monitoring, but unusually high scores prompted an expert review and suspended admissions.
Mexico's National Autonomous University, or UNAM, has ordered approximately 58,000 undergraduate applicants to take a new in-person entrance exam after an investigation found evidence of suspected widespread cheating in the university's first fully remote admissions test.
According to The Guardian, the repeat test applies to applicants who were provisionally admitted after the 2026 exam. Ars Technica reports that it also covers applicants who would have met the minimum successful score for their chosen program under admission thresholds used from 2021 through 2025. UNAM's admissions decisions will depend on the new test results.
The examination was taken by nearly 160,000 applicants over several weeks from late May through early June, Ars Technica reports. CNN reported that UNAM suspended enrollment while it investigated irregularities, leaving applicants uncertain ahead of the semester scheduled to begin on August 10.
Score anomalies triggered the review
The university appointed an expert commission after results showed a sharp rise in high scores. Ars Technica and Heise report that 16.3% of applicants scored at least 100 out of 120 in 2026, compared with an average of 3.5% between 2021 and 2025. The share scoring 110 or more rose from 0.9% in the earlier period to 5.5% this year.
Heise reports that the commission suspected cheating in roughly half of the nearly 160,000 exams, while around 2% had already been formally annulled for rule violations.
Those figures do not establish that every affected applicant cheated. The broad retest population includes students whose scores were not individually invalidated, and CNN reported that some applicants have protested the decision. According to Heise, UNAM rector Leonardo Lomeli Vanegas apologized for requiring candidates to prepare for another exam and described the retest as necessary for equal opportunity.
Remote proctoring did not prevent suspected workarounds
For the online exam, UNAM used Respondus LockDown Browser, intended to restrict access to other applications and websites during testing, alongside AI-assisted webcam monitoring from Territorium, according to Ars Technica, Gizmodo, and Heise. Heise reported that human supervisors reviewed candidates' video feeds with AI assistance.
Heise also reported that instructions circulated before the exam advising applicants to use second screens outside the camera's field of view and conceal earpieces, while test questions were reportedly offered for sale during the 19-day testing window. The outlet reported that UNAM filed criminal complaints related to potential cheating offers before the commission published its findings.
The episode is a consequential reminder that browser lockdown and webcam-based proctoring constrain only the devices and visual environment they can observe. In comparable high-stakes remote-testing systems, external devices, leaked content, identity verification, and adjudication of anomalous behavior remain separate control problems. AI-based monitoring can generate signals for review, but it cannot by itself validate the integrity of an entire examination cohort.
For ML and data practitioners building assessment or proctoring systems, the unusually large score distribution shift is the central operational finding. A monitoring stack should be evaluated not only on whether it flags individual violations, but also on whether its telemetry, audit process, and test-security design can support defensible admissions decisions when aggregate outcomes depart sharply from historical baselines.
Key Points
- 1UNAM ordered roughly 58,000 applicants to retake an in-person exam after anomalous online-test results and a review of suspected cheating.
- 2Applicants scoring 100 or more rose from a 3.5% historical average to 16.3%, making distribution monitoring central to the investigation.
- 3AI webcam monitoring and locked browsers cannot independently control external devices, leaked questions, or cohort-level integrity.
Scoring Rationale
The case is a notable real-world failure involving AI-assisted proctoring in a high-stakes admissions workflow affecting tens of thousands of candidates. It offers practical evidence that endpoint controls and behavioral monitoring need robust test-security, auditing, and escalation processes, though it is not a new model or broadly deployed developer platform.
Sources
Public references used for this report.
Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Ad Tech problems
