my timesThe Korea Times

REPORTER’S NOTEBOOKAI cheating is forcing Korea to rethink CSAT

Listen

Security fixes may buy time, but technology is accelerating debate over purpose of testing

An official tries on artificial intelligence smart glasses during a training session for College Scholastic Ability Test staff at the Seoul Metropolitan Office of Education in Seoul, Nov. 11. Yonhap

An official tries on artificial intelligence smart glasses during a training session for College Scholastic Ability Test staff at the Seoul Metropolitan Office of Education in Seoul, Nov. 11. Yonhap

Ten seconds. That was all it took for artificial intelligence (AI) glasses to solve four college entrance exam-style math questions in a recent demonstration by The Korea Times.

For Korea’s make-or-break annual College Scholastic Ability Test (CSAT), the immediate question is how to keep that kind of assistance out of the exam room. The larger one is whether keeping AI out would resolve a more fundamental issue: what students should be learning in an AI-driven world.

In the short term, finding solutions to the first question is crucial to keeping the exam fair and credible. The recent tax-accountant examination cheating case shows that one can use AI with a miniature camera and smartphone to obtain answers throughout an exam without being detected. Had the cheater’s unusually high score not drawn scrutiny, the AI cheating may have gone unnoticed.

The setup did not even involve AI glasses — the devices now drawing education authorities’ attention ahead of Thursday’s CSAT.

The next generation of devices could make detection harder still. Smart contact lenses are under development in Dubai and Japan, with some consumer applications targeted for 2030. Authorities may eventually have to look beyond the glasses students wear to the lenses on their eyes.

Blocking internet access may not be enough either. On-device AI, such as Google’s Gemma 3n, can process images and text without connecting to an external server. This means that even internet restrictions cannot prevent its use.

Research prototypes also show how commands could be understood even without speaking aloud. Cornell’s EchoSpeech recognizes a limited set of silently mouthed commands using sensors on glasses, while InvisiSpeech uses a sensor mounted around a back molar to detect subtle vibrations inside the mouth.

These are not ready-made cheating systems. But they point to a future in which AI becomes harder to see, hear or block.

Even a perfectly secured test without cheating would leave its educational purpose open to question. Independent problem-solving still matters. Students need knowledge of their own to judge whether an AI-generated answer is accurate or evidence-based. But that is also a reason to give greater weight to explaining a solution, spotting a less obvious error and applying knowledge to an unfamiliar problem — not simply selecting the right answer under time pressure.

Increasing calls for CSAT reform reflect that concern. Seoul education superintendent Jung Geun-sik has proposed moving toward absolute grading and written-response assessments, with the eventual abolition of the CSAT for 2040 admissions.

Growing concerns about AI cheating and a shrinking student population strengthen the case for accelerating reform. This year’s elementary-school intake fell below 300,000 for the first time. With four-year universities alone offering over 350,000 admission slots, demographic change will force a reckoning by the time these children reach college age.

As past attempts have shown, reform will not be easy. Any alternative must earn the trust of students and parents rightly concerned about admissions fairness, and consensus on what should replace the current system remains elusive. But by calling the future of standardized testing into question, AI may force a debate that has been delayed for too long.

Explore More

  • Q.

  • Q.

  • Q.