UNAM Forces 58,000 Students to Retake Entrance Exam After AI Proctoring Fails

Remote Testing Experiment Produces Suspicious Results

Nearly 160,000 applicants took the entrance exam for Universidad Nacional Autónoma de México (UNAM) this summer in the institution's first fully remote administration of the high-stakes test. The experiment with AI-powered proctoring has backfired spectacularly: the university will now require approximately 58,000 students to retake the exam in person after results showed statistically improbable score inflation.

Between 2021 and 2025, 3.5 percent of test-takers scored 100 or higher on the 120-question exam. This year, that figure jumped to 16.3 percent. At the highest performance tier, the pattern was even more dramatic — the percentage of students scoring 110 or above increased from 0.9 percent to 5.5 percent.

According to Ars Technica, which first reported the findings, UNAM convened an expert commission to investigate the anomalies. The commission recommended a "control exam" administered in person to restore confidence in the admissions process.

AI Supervision Proved Insufficient

The remote exam relied on multiple security layers. Test-takers used LockDown Browser from Respondus, which prevents printing, copying, accessing other websites, or switching to different applications during the exam. UNAM also deployed AI proctoring software from Territorium that monitored webcam feeds for signs of cheating, including detecting if someone else replaced the registered applicant, spotting cell phones or earphones, or flagging when test-takers left the camera frame.

One human supervisor monitored every 150 applicants, receiving alerts from the AI system about potential irregularities. Despite these measures, the safeguards appear to have been inadequate. UNAM canceled nearly 2 percent of exams for conduct violations, but the statistical evidence suggests far more widespread problems.

AI expert Raul Rojas told NPR that statistical modeling indicated almost half of students may have cheated during the online exam. The New York Times reported that cheating strategies circulated widely before the test, including advice to position monitors outside the webcam's view to access ChatGPT, hide earphones under hair, or even hire proxy test-takers who stayed off camera.

Why It Matters

This incident exposes critical vulnerabilities in AI-based remote proctoring at scale, particularly for high-stakes assessments. As educational institutions worldwide continue evaluating remote testing options, UNAM's experience demonstrates that current AI supervision technology cannot reliably prevent determined cheating when students control the testing environment. The failure carries immediate consequences for tens of thousands of students who must now re-take the exam, and broader implications for universities considering similar approaches. The case also highlights how multiple-choice formats may be especially vulnerable to remote cheating compared to essay-based assessments where AI-generated content is more detectable.

Return to In-Person Testing

The control exam will affect not only students who secured admission based on this year's scores, but also those who would have qualified under the minimum successful scores from 2021 through 2025. UNAM's rector apologized to honest applicants who must now prepare for and take another test despite doing nothing wrong, but called the measure necessary to ensure fairness and certainty in the admissions process.

Classes are currently scheduled to begin August 10, requiring rapid implementation of the new testing protocol. Details about the control exam are expected to be announced soon, according to Gaceta UNAM, the university's official publication.

The details of this incident were first reported by Ars Technica.