AI Cheating in Exams: South Korea’s Top Scorer Admits It
Key Takeaways
- The top scorer on South Korea's national tax accountant exam has admitted cheating with generative AI, after first denying the allegations.
- The candidate, a 34-year-old known as Han, scored 397.5 out of 400 in the first round, missing only one of 160 questions.
- He said he photographed questions with a microcamera, sent them to an AI tool, and checked the answers on a smartphone.
- He has submitted a written confession to Seoul police, who have opened an investigation, and the exam body is moving to revoke his score.
- The confession came days before South Korea's national college entrance exam, prompting an emergency government review of suspected AI cheating.
- The case exposes how far exam security has fallen behind the technology candidates can now carry into a hall.
A Near-Perfect Score That Nobody Believed
Some results are so good that they invite suspicion rather than admiration. That is what happened to the top scorer on South Korea's national tax accountant exam. He finished first in both rounds of the qualification exam, held in April and July 2026, and his first-round score of 397.5 out of 400 points looked, to many observers, almost too good to be true. Out of 160 questions, he had answered all but one correctly.
The scepticism began almost as soon as he shared his results online. According to the Korea JoongAng Daily, he posted his scores on a Wednesday: 97.5 in public finance in the first round, and a perfect 100 in introduction to tax law, introduction to accounting, and commercial law. Within days, accusations were circulating that he had cheated using AI smart glasses. He denied them.
On Sunday 15 November, the sceptics were proved right. The candidate, identified by the BBC as Han Won-seok, 34, and known publicly only by his surname, Han, appeared on the YouTube channel of Attorney Lee Don-ho, the lawyer representing him, wearing a face mask. There, he admitted what he had denied. "I cheated using electronic devices and generative AI. My passing the exam with the top score was not the result of fair effort," he said, as reported by the Guardian and the Korea JoongAng Daily.
The Guardian's report, published on 15 November under the headline "160 questions, only one wrong: top scorer in South Korea tax exam admits AI cheating", set out the confession in plain terms. It was not a marginal case of looking up a single answer. Han said he had cheated in both rounds of the exam, using the same method each time. He also said, according to the Seoul Economic Daily, that he had used the same method in the first round of the certified public accountant exam.
The timing could hardly have been worse for South Korea's exam authorities. The confession landed just days before the country's annual college entrance exam, the CSAT, one of the most high-stakes tests in the world, in which hundreds of thousands of teenagers sit the same paper on the same day. The presidential office moved within hours.
Which Exam, and Why It Matters
The exam at the centre of the scandal is the national tax accountant qualification exam, the 63rd sitting of the test, administered by the Human Resources Development Service of Korea, known as HRDK. It is the gateway to becoming a certified tax accountant in South Korea, a profession that carries real status and earning power. Passing it is difficult. Topping it, with a near-perfect score, is the kind of result that opens doors for a career.
That is part of why the case has hit so hard. This was not a school test or a low-stakes quiz. It was a national professional qualification, run by a state body, with invigilators in the room. And the cheating, if Han's account is accurate, happened in plain sight. According to the Seoul Economic Daily, Han sat in the front row, directly in front of the podium at a test site staffed with two proctors, and was not caught at the scene.
The exam body has now moved against him. HRDK said after an emergency meeting that it had decided to file a complaint against Han for obstruction of business and to proceed with revoking his score, the Korea JoongAng Daily reported. The Korean Association of Certified Public Tax Accountants said it would launch a new evaluation system during practical training for apprentice tax accountants, designed to screen out people who passed by cheating.
There is a small but important point of confusion in the reporting that is worth flagging honestly. Some outlets describe the exam as the tax accountant exam, while others, including the Guardian's own article, refer at points to the certified public accountant exam. The two are different qualifications in South Korea, and Han is reported to have sat both. The confession itself, and the score of 397.5 out of 400, relate to the tax accountant exam. Readers should treat any single label with care until the exam body publishes a fuller account.
How the Cheating Worked
Han's own description of the method is strikingly simple, and that simplicity is the point. In the video posted by his law firm, he said: "I took photos of the exam questions with a micro camera, sent them to AI and then checked the answers on a smartphone." The Korea JoongAng Daily quoted the same account, and the Guardian reported the same three steps: photograph the questions with a microcamera, send them to an AI tool, and check the answers on a smartphone.
Nothing in that description requires exotic technology. A tiny camera, a phone, and access to a generative AI tool are enough. That is what makes the case so uncomfortable for exam authorities everywhere. The tools involved are ordinary consumer technology, widely available and cheap, and the AI models that can read a photographed question and answer it are now good enough to handle professional exam material in seconds.
What has not been reported, at least in the coverage available so far, is which AI tool Han used, how he concealed the phone, or how long the answers took to come back. Those details matter, because they determine how detectable the method is. A phone checked openly would be spotted. A phone checked under a desk, in a toilet break, or through a wearable device would be much harder to catch. Until the police investigation produces more detail, those questions remain open, and this article does not fill them in with guesses.
What is known is that the method was not detected on the day. Two supervisors were present when the candidate sat the test, HRDK confirmed, and he was not stopped. The cheating was exposed only after the results themselves drew suspicion, and after Han chose to confess. That sequence, detection by suspicion rather than by invigilation, is the detail that should worry every exam board.
The Confession and the Police Investigation
Han did not only confess on video. He also submitted a written confession to Seoul police through his lawyer. The Seoul Metropolitan Police Agency confirmed that it had received the written confession and said it had assigned the case to investigators, specifically its public crime investigation team, according to the Korea JoongAng Daily.
The legal exposure is serious. Article 314 of South Korea's Criminal Act states that anyone who obstructs another's business by deception or force faces up to five years in prison or a fine of up to 15 million won, which the Korea JoongAng Daily converted to about 11,300 US dollars. HRDK's complaint against Han is framed as obstruction of business, the same category.
Alongside the criminal process, the administrative consequences are already taking shape. The BBC reported that the government said those found to have used AI to cheat on the tax accountancy exam will have their qualification revoked and be banned from retaking the test for five years. The Seoul Economic Daily reported the same direction of travel: the government is reviewing whether to suspend the validity of the result and is moving to revoke the pass and bar Han from taking the exam for five years under the Certified Tax Accountant Act.
None of this is final yet. The revocation is described as in progress, not complete, and the police investigation is at an early stage. But the direction is clear, and the severity of the proposed penalty, a five-year ban, signals how seriously South Korea treats the integrity of its national qualifications.
Why the Timing Matters: The CSAT Is Days Away
The confession arrived at the worst possible moment for South Korea's exam system. The national college entrance exam, the CSAT, is scheduled for Thursday 19 November, and it is an event of extraordinary national importance. On the day of the exam, much of the country reorganises itself around the test: flights are delayed, offices open late, and police escort late students to their exam halls. Hundreds of thousands of teenagers sit the same paper, and the result shapes university admissions and, in the eyes of many families, the course of a life.
It was against that backdrop that the presidential office acted. Ha Joon-kyung, the presidential director of national policy, convened an emergency meeting on Sunday, where officials agreed to look into every case flagged so far, the Korea JoongAng Daily reported. Prime Minister Han Seong-sook acknowledged public concerns over AI-assisted cheating, according to the Guardian's coverage.
The response is not limited to the tax exam. The presidential office ordered a sweeping review of suspected AI cheating across national qualification exams, and for the college entrance exam itself, the authorities will put in place AI device screening procedures for all test-takers and deploy every available method, including thermal imaging cameras, the Seoul Economic Daily reported. Thermal imaging is designed to detect electronic devices concealed on a person, including phones and wearable devices that a visual check or a metal detector might miss.
Whether those measures will be enough is an open question. Screening every candidate with thermal cameras is a major logistical exercise, and it is being assembled in days, not months. The deeper problem is that the technology keeps shrinking. A microcamera is, by definition, hard to spot, and a smartphone can be hidden in ways that no single screening method reliably defeats.
This Was Not an Isolated Case
One of the most important details in the coverage is that Han's case is not the only one. According to data HRDK submitted to the office of Representative Kim Wi-sang of the People Power Party, five cases of cheating with smart glasses were caught on site through October this year at national qualification exams administered by the agency. That figure covers only cases that were caught in the act. It says nothing about how many went undetected.
The pattern is consistent with what exam authorities in other countries have been reporting for the past two years. Smart glasses, which can photograph a question and display an answer in the wearer's field of vision, have become a recurring concern. Microcameras and earpieces have been around for longer. What has changed is the AI layer: where a cheating candidate once needed a confederate outside the hall to look up answers, a generative AI tool can now do the job instantly, with no second person involved.
That shift matters because it removes the weakest link in the old cheating chain. There is no accomplice to catch, no phone call to intercept, no second person whose behaviour might give the game away. The entire operation can be carried out by one person, with devices that are small enough to hide and an AI service that answers in seconds.
What Exam Bodies Can Actually Do
The uncomfortable truth in this case is that detection failed at every stage. The candidate was not caught in the hall, not flagged by the invigilators, and not stopped by whatever screening was in place. He was caught, in effect, by his own score. That raises a hard question for exam bodies: if a near-perfect result is the only reliable signal, how many less spectacular cheats are going unnoticed?
There are no easy answers, and this article does not pretend otherwise. But the options being discussed fall into a few broad groups, and each has real trade-offs.
The first is better screening at the door. South Korea's plan for the CSAT, AI device screening for all test-takers plus thermal imaging, is the most visible example. Metal detectors, signal jammers, and device bans are already used in many exam systems. The problem is that each measure has a countermeasure, and each adds friction and cost. Jamming a signal in a hall full of candidates is legally restricted in many countries. Thermal imaging works best when operators know what they are looking for.
The second is changing what exams test. If a question can be answered by photographing it and sending it to an AI tool, then the question is, in a sense, testing something AI can already do. Exams that test reasoning under supervision, applied judgement, or work produced over time are harder to cheat on in this way, but they are also harder and more expensive to run at national scale. South Korea's CSAT, like most national exams, is built around standardised questions that can be marked at scale. Redesigning that system is a project of years, not weeks.
The third is detection after the fact. AI detectors, which claim to identify AI-generated text, are widely used in education, but their accuracy is contested. Our guide to how accurate AI detectors are in 2026 sets out the limits in detail, and the evidence is not reassuring: detectors can produce false positives, and they are unreliable on short answers or heavily edited text. For more on the specific tools involved, see our review of how accurate Turnitin AI detection is in 2026, which examines the detection system used by many universities. In an exam hall, where the output is a set of multiple-choice answers rather than an essay, detectors are of little use at all. A correct answer looks the same whether it came from a candidate's knowledge or from an AI tool.
The fourth option, and perhaps the most honest one, is to accept that some cheating will happen and to focus on making the consequences severe and certain enough to deter it. That appears to be the direction South Korea is taking, with a five-year ban and a criminal complaint. Deterrence only works, however, if candidates believe they will be caught. A case that was exposed by suspicion and confession, rather than by detection, sends a mixed message on that point.
What This Means for Students, Teachers, and Everyone Else
For the honest candidates who sat the same exam as Han, the damage is real and hard to undo. A qualification loses value when the public doubts whether it was fairly earned. The Korean Association of Certified Public Tax Accountants' decision to add a new evaluation stage during practical training is an attempt to repair that trust, but it also imposes a cost on every future candidate, including those who never cheated.
For students more broadly, the case is a warning about the line between using AI to learn and using AI to cheat. The distinction matters. Using an AI tool to understand a concept, practise a problem, or check your reasoning is learning. Using it to obtain answers during a closed-book exam is cheating, and the consequences, as Han is discovering, can include a criminal record and a five-year ban. The technology is the same in both cases. The context is what changes the meaning.
For teachers and exam designers, the case is a prompt to rethink what assessment is for. If AI can answer the questions, then the questions may be testing the wrong thing. That is a harder conversation than buying better detectors, but it is the one that ultimately matters. Australia's approach to generative AI in schools, which we examined in our guide to Australia's generative AI in schools framework, takes a different route: setting out when and how AI may be used, rather than pretending it can be kept out entirely. Different countries will reach different answers, but the question is now unavoidable everywhere.
For the rest of us, the case is a reminder that AI cheating is no longer a story about students copying essays. It is a story about professional qualifications, national exams, and the integrity of the systems that decide who gets to practise a profession. The tools are cheap, the method is simple, and the detection is hard. That combination is not going away.
What We Could Not Verify
- The Guardian's website could not be opened directly during research for this article, so the Guardian's account is reported here through the newspaper's own published summary and through secondary reporting that quoted it. The headline, the score, and the quoted confession are consistent across multiple independent outlets.
- Which generative AI tool Han used has not been named in the reporting reviewed. No article identified the specific model or service.
- How the phone was concealed, and how answers were checked without being seen, has not been described in the reporting reviewed.
- Whether Han's confession covers only the tax accountant exam or also the certified public accountant exam in full: the Seoul Economic Daily reported he said the same method was used in the first round of the certified public accountant exam, but the scope of that admission is not fully detailed in the coverage available.
- The exact number of candidates affected, and whether anyone else cheated on the same exam using the same method, has not been established. The five smart-glasses cases reported by HRDK are the only confirmed figure.
- The revocation of Han's score and the five-year ban are reported as in progress or planned, not as completed decisions.
- The precise screening measures planned for the CSAT, beyond AI device screening and thermal imaging cameras, have not been published in detail.
Sources
- The Guardian, "160 questions, only one wrong: top scorer in South Korea tax exam admits AI cheating", 15 November 2026: https://www.theguardian.com/technology/2026/nov/15/ai-cheating-south-korea-college-entrance-exam
- Korea JoongAng Daily, "AI cheating confession prompts Korea review of national exams before CSAT", 16 November 2026: https://www.koreajoongangdaily.com/korea/ai-cheating-confession-puts-koreas-exams-under-scrutiny-ahead-of-csat/12974040
- Seoul Economic Daily, "AI Glasses Cheating Scandals Shake Korea's National Exams", 16 November 2026: https://en.sedaily.com/society/2026/11/16/ai-glasses-cheating-scandals-shake-koreas-national-exams
- Are AI Detectors Accurate? What 2026 Research Shows
- How Accurate Is Turnitin AI Detection in 2026?
- Australia's Generative AI in Schools Framework: What It Actually Says