AI exam proctoring went wrong at Mexico's biggest university, and 58,000 students must sit the test again
Nearly 160,000 applicants took UNAM's entrance exam remotely this year, watched by AI software instead of human invigilators. The results looked nothing like any previous year, and now tens of thousands of students face a retake.
Key points
- Around 160,000 students sat UNAM's entrance exam remotely for the first time in May to June 2025, monitored by AI webcam proctoring software.
- In previous years (2021 to 2025), 3.5 percent of test-takers scored 100 or more out of 120 questions; this year that figure jumped to 16.3 percent.
- The spike in top scores triggered an investigation, and approximately 58,000 students must now retake the exam.
- UNAM is Mexico's largest university, and its entrance exam is one of the most competitive in Latin America.
Mexico's biggest university, UNAM (Universidad Nacional Autónoma de México), tried something bold this summer. It moved its famously competitive entrance exam entirely online, with about 160,000 applicants sitting the test from their own homes between late May and early June 2025.
Instead of human invigilators in an exam hall, the university used AI-powered webcam proctoring software, the kind that watches your face through your laptop camera and flags anything that looks suspicious, combined with a "lockdown" browser that blocks you from opening other tabs or searching the web.
It did not go well.
What went wrong with the results?
The scores that came back looked nothing like any previous year. Between 2021 and 2025, around 3.5 percent of students scored 100 or more on the 120-question test. This year, 16.3 percent did. That is roughly five times as many top scorers as normal, in a single sitting.
To put that in plain terms: if your child's school suddenly went from two A-grades per class to ten, you would want to know why.
The jump was big enough that UNAM launched an investigation. As first reported by Ars Technica, the conclusion was damaging: the remote setup had not kept the exam secure, and results could not be trusted. About 58,000 students now have to sit the test a second time.
What does this mean for the students affected?
For the 58,000 students called back for a retake, this is genuinely disruptive, especially for anyone who cleared space in their schedule, arranged childcare, or travelled to sit the original exam.
More broadly, it raises a real question about whether AI proctoring software is ready to guard a high-stakes test on its own. The technology watches for things like looking away from the screen or a second voice in the room. But determined students, and a 16 percent top-score rate suggests a lot of them found a way, can apparently work around it.
Should other universities worry?
Yes, and many already do. AI proctoring, software that monitors students remotely during online tests, has grown fast since the pandemic pushed exams off campus. The pitch is simple: cheaper than renting a hall, easier for students who live far away.
The catch is that the software has a mixed record. It can flag innocent behaviour (looking sideways to think) as cheating, and miss actual cheating that happens off-camera.
UNAM's case is an unusually large and public example of what can go wrong. A retake for 58,000 people is a significant cost, in money, time, and trust.
For students anywhere facing an AI-proctored exam: read the rules carefully, check your camera angle and lighting beforehand, and keep your phone in another room. Not because the software is always fair, but because a false flag is a headache you do not need.
Common questions
Is AI proctoring used at universities outside Mexico?
Yes. Tools like Respondus, ProctorU and Honorlock are used at universities across the US, UK, Canada and Australia, often for online courses and remote entrance tests.
Does AI proctoring software actually stop cheating?
The evidence is mixed. Studies show it catches some rule-breaking, but students can obscure notes off-screen or use a second device outside the camera's view. UNAM's results suggest determined students found workarounds at scale.
Can a student appeal an AI proctoring flag?
Policies vary by institution, but most universities allow a human review of flagged footage. If you are ever flagged, ask immediately for a human to review the evidence rather than accepting an automated decision.
