COVER STORY AI glasses solve 4 CSAT math questions in 10 seconds

At an optical store in Seoul’s Myeong-dong shopping area, I looked through a pair of Rokid artificial intelligence (AI) glasses at a page of four math questions from a mock version of the College Scholastic Ability Test (CSAT).

Within 10 seconds, answers and explanations appeared inside the lenses, before I had even finished making sense of the first question. All four were correct.

I then tested the glasses with another math question — this time a calculus problem involving a definite integral. The correct answer appeared in under two seconds.

“This is the very product that has been in the controversy lately over cheating,” the optician said.

It was a brief demonstration, not a controlled accuracy test. But reading the solutions inside the lenses, I could see why exam authorities were concerned ahead of this year’s CSAT, set to be held on Nov. 19.

Data from the Human Resources Development Service of Korea, disclosed in October during a parliamentary audit, showed that five cases involving AI smart glasses had been detected at state-run qualification examinations this year. Four involved Chinese-manufactured Rokid products and one involved glasses from U.S. tech giant Meta.

The appeal to a would-be cheater was easy to understand. The Rokid glasses combined a camera to capture questions and a display to show AI-generated answers, accessed through a connected smartphone.

The glasses retail for 999,000 won ($750) in Korea, but buying a pair is not the only option. One rental service advertises rates as low as 9,900 won a day.

I also tried a pair of Ray-Ban Meta glasses. Unlike the Rokid model, the pair I tested had no display. Instead, I spoke to it and listened to replies through its built-in speakers. Meta sells a separate display-equipped model in the U.S., but it is not officially available in Korea. The audio exchange felt more conspicuous and riskier from a cheater’s perspective than reading text inside the Rokid lenses, though that hardly rules out potential for misuse.

The Rokid glasses had limitations, too.

At one point, the system indicated that it could not see the question clearly and needed a better view. At other times, it stopped responding and appeared to wait for another order rather than continue following my earlier instructions.

Other questions remained unanswered. Would an instruction like “show only the correct answers, without explanations, for the next 30 minutes” remain effective throughout the entire period? How long would the battery last under repeated use? I did not test either, but the speed of the demonstration suggested that even brief use could pose a cheating risk.

The glasses themselves, however, could give a cheater away. The frames are thicker than ordinary eyewear and the Rokid lenses have a noticeable and distinct reflection, but spotting those differences in a crowded exam room could be harder than noticing them in a shop.

My brief trial did not prove that the glasses could get someone through an exam undetected. It did, however, show how quickly they could deliver correct answers — and why exam authorities face a difficult task.