AI-proctored exam glitch forces 58,000 Mexican students to retake entrance test
UNAM's first fully remote, AI-supervised admissions exam produced wildly inflated top scores — so the university is making tens of thousands retake it in person.
What happened: Nearly 160,000 people took UNAM's entrance exam this spring, the first time Mexico's largest university ran it fully remotely, using a lockdown browser plus AI webcam proctoring. When scores came back, they were bizarre: 16.3% of takers scored 100+ out of 120, versus a 2021-2025 average of 3.5%. Scores of 110+ jumped from 0.9% to 5.5%. UNAM convened an expert commission, which just recommended an in-person 'control exam' retake affecting roughly 58,000 people — not just this year's admits but everyone who cleared minimum scores by program since 2021.
Why it matters: This is one of the largest real-world stress tests yet of AI exam proctoring, and it seems to have failed at real scale rather than in a pilot. A jump in top scores this dramatic points to widespread, largely undetected cheating, undermining the exam's core purpose: fairly ranking applicants for scarce university spots. Now honest students who did nothing wrong must re-prepare and retest, on a tight clock before classes begin August 10.
How it works, plainly: Lockdown browsers block access to other apps and tabs during a test, while AI proctoring software watches webcam feeds for suspicious behavior — eyes wandering, extra voices, phones in frame. In theory this substitutes for a human invigilator in a room. In practice, it depends on the software reliably flagging cheating methods candidates devise, from hidden earpieces to off-screen helpers to workarounds nobody anticipated. UNAM's results suggest whatever it deployed missed a lot.
The rollout: UNAM's rector has apologized to applicants who tested honestly but must now sit for the control exam anyway, calling the retake necessary 'to give certainty and guarantee equity in access.' Details of the new in-person test are still being finalized, and the university may need to delay the academic year if it can't organize testing for 58,000 people before classes start. The episode will likely shape how — or whether — UNAM and other large institutions use remote AI proctoring again.