When nearly 160,000 people took the entrance exam for Mexico’s UNAM university this year, they did so from their own homes for the first time. To keep things fair, the school used specialized software to lock down the students' computers and AI algorithms to watch them through their webcams. The results were record-breaking, but in the wrong way: the number of perfect scores skyrocketed, suggesting that the digital safeguards failed to stop mass cheating.
The university discovered that students were scoring at the highest levels at rates far exceeding historical data. After investigating, officials concluded that the remote setup was likely compromised. The university decided the only way to ensure fairness was to force roughly 58,000 applicants to retake the exam in a physical room under the watch of human proctors. This is a massive logistical hurdle that impacts thousands of innocent students who now have to prepare for the test all over again because the digital security measures proved ineffective.
The limits of digital oversight
The university relied on two layers of technology. First, a lockdown browser blocked students from opening other tabs, copying text, or accessing outside apps on their computers. Second, AI proctoring software analyzed webcam feeds. This AI is trained on vast amounts of data to recognize patterns of suspicious movement, like someone looking off-screen, a secondary person entering the room, or a phone being used nearby. When the software detected these patterns, it flagged the event for a human supervisor.
However, these systems are essentially trying to catch the future by looking for the past. They are only as good as their ability to recognize specific, pre-programmed signs of cheating. If a student finds a way to cheat that the AI wasn't specifically trained to spot—like placing a monitor just outside the camera's view or using tiny, hidden earpieces—the system is essentially blind. The AI sees a student sitting still and staring at the screen, and to its logic, that looks exactly like someone taking a test honestly.
This incident reveals a fundamental gap in how we use AI to police our world. We often treat AI like an omniscient judge, but it is actually a pattern-matching machine. It relies on the assumption that cheating will look a certain way. When applicants realized this, they moved their cheating tactics into the blind spots of the technology. The result was not just a security failure, but a reminder that as we move more high-stakes human processes online, we risk trading deep, reliable human judgment for a false sense of security provided by algorithms that are easily outmaneuvered.
Liked this one? The next lands at breakfast.
Every story in tomorrow's AI news, rebuilt in plain English — five minutes, sources linked, free forever.
By joining you agree to receive Article's daily newsletter — unsubscribe in one click. Privacy