UNAM used AI-assisted monitoring to protect the integrity of its first fully online admissions exam, but suspected cheating, leaked questions, and abnormal scores have now forced tens of thousands of applicants back into an in-person test.
WHAT’S HAPPENING
Mexico’s National Autonomous University, UNAM, moved its highly competitive undergraduate admissions exam online in 2026 in an effort to make access easier for applicants across the country.
To protect the process, the university used identity verification, environmental monitoring, human supervisors, and AI-supported systems designed to flag suspicious behavior.
UNAM said the AI itself did not make automatic decisions to cancel exams. Instead, it generated alerts that university staff reviewed.
But after the test, unusually high scores, evidence of impersonation, outside assistance, leaked questions, and suspected AI-assisted cheating raised serious concerns about the integrity of the admissions process.
The university ultimately ordered roughly 58,000 applicants to take an additional in-person control exam.
WHY IT MATTERS
This is not simply a story about students finding ways to cheat.
It is a test of whether institutions can rely on AI-assisted surveillance to secure high-stakes digital processes.
UNAM’s system was designed to detect suspicious behavior, but the broader testing environment still created vulnerabilities that technology alone could not eliminate.
That exposes an important limitation:
AI monitoring can identify signals, but it cannot automatically guarantee trust.
High-stakes systems still depend on secure processes, human review, identity verification, controlled environments, and institutional judgment.
WHO BENEFITS
Universities and testing organizations gain a real-world warning about the limits of remote AI proctoring.
Security and identity-verification companies may see increased demand for stronger systems combining AI monitoring with better authentication and fraud detection.
Students who legitimately earn admission benefit when institutions strengthen controls designed to protect fairness.
Education policymakers gain another case study showing that accessibility and exam integrity have to be designed together.
WHO LOSES
Students who followed the rules may now have to repeat an exam because of suspected misconduct by others.
Universities relying heavily on remote testing face higher reputational and operational risks if their systems cannot reliably distinguish legitimate behavior from fraud.
AI proctoring vendors may face greater scrutiny over what their technology can realistically prevent versus merely detect.
Institutions also risk losing public trust if automated monitoring appears intrusive while still failing to stop widespread cheating.
WHAT HAPPENS NEXT
High-stakes remote testing is likely to become more cautious.
Universities may combine AI surveillance with stronger identity verification, secure browsers, controlled testing centers, random in-person verification, and more aggressive fraud investigations.
The broader lesson goes beyond education.
As governments, companies, and schools use AI to supervise increasingly important digital processes, they will have to recognize a basic reality:
Automated oversight does not create trust by itself.
The system surrounding the AI still determines whether the result can be trusted.
