What happened
Hankyung reported on November 18, 2026, that Seoul Metropolitan Police Agency investigators had requested an exit ban for Han Won-seok the previous day and that the Ministry of Justice approved it on November 18. The restriction prevents him from leaving South Korea for six months. The report says Han admitted using generative AI and electronic devices to cheat in the first- and second-stage sessions of the 63rd tax accountant examination, held in April and July.
The tax accountant examination is a professional qualification test, and Hankyung identifies Han as its top scorer. The reported conduct is therefore not framed as a hypothetical misuse of a consumer AI service. It is the subject of an active police investigation into alleged misconduct in a consequential credentialing process. Hankyung says the police sought the travel restriction after questioning Han and that the Ministry of Justice approved the request. The article does not state whether prosecutors have filed charges or whether a court has made findings on the allegations.
According to the article, Han told police that he attached a miniature camera to correction tape, photographed exam questions, and viewed answers generated by AI on a miniature liquid-crystal display. The report also says a modified calculator was used. Calculators were among items permitted in the examination room, making the reported modification notable: the device’s allowed category did not ensure that its altered form was harmless. Hankyung does not identify the AI product, describe the network connection, or explain whether the calculator transmitted information, displayed it, or served another role.
Hankyung says Han voluntarily submitted a miniature camera, mobile phone, and laptop to police after turning himself in. Investigators questioned him for about 13 hours on November 17. On November 18, police searched his residence for evidence that he had not submitted. Those actions indicate that investigators were seeking to establish the scope and mechanics of the alleged cheating rather than relying solely on the reported admission. The article characterizes the inquiry as continuing.
The same article also discusses separate allegations involving Lee Jin-woo, a top scorer in the Korean Law School Admission Test, or LEET. Marks on a publicly shown answer sheet led to speculation that answers had been changed after being entered one position off. Private academy Sidae Injae said it would terminate its contract with Lee while facts were checked, and complaints were filed with police. Those allegations concern a different examination and different person. Hankyung does not present them as established findings, and they should not be conflated with Han’s reported admission.
What the archive adds
New Runtime’s earlier coverage reported that South Korea had launched a task force over misconduct in the tax accountant exam and begun procedures to invalidate an offender’s result (November 20, 2026). The travel ban is a new development in that same case, adding a police measure and details of the reported device setup rather than opening a separate story.
Trend signal
This does not yet fit a pattern New Runtime tracks. The report concerns alleged examination fraud and law-enforcement action, not a demonstrated shift in how AI systems are built, governed, or deployed. Its relevance is narrower: it is a concrete example of an AI system being joined to a physical capture-and-return channel in a controlled setting.
Why now
The reported travel ban follows an admission, voluntary submission of several devices, and a lengthy police interview. Hankyung says police then searched Han’s residence for further evidence. The timing therefore reflects an investigation moving from a reported confession to evidence collection and a measure intended to keep the suspect available while that work continues.
Why it matters
For teams that design AI-enabled assessments, the reported method is a reminder that integrity controls cannot focus only on what software runs on a test-taker’s device. A camera can extract the questions, an AI system can generate responses elsewhere, and a concealed display can return them to the person taking the test. The resulting chain crosses physical objects, software services, communications links, and human action.
The reported use of a modified, normally permitted calculator adds a second lesson. A rule that permits a class of object is not necessarily a control over what a particular object can do. Assessment operators need procedures that distinguish ordinary permitted equipment from altered equipment, while also avoiding disproportionate searches or assumptions about every candidate.
The article does not establish how the equipment operated or whether the reported method could be reproduced at scale. Still, it points toward layered controls: assess the physical environment and device capabilities alongside proctoring, question design, access policies, and post-exam review. No single prompt, declaration, or software restriction can secure an assessment once questions can leave the room through an unobserved physical channel.
Limits and uncertainty
Hankyung reports Han’s admission, the police investigation, and the travel ban, but the article does not establish the case’s final legal outcome. It does not provide technical specifications for the camera, display, modified calculator, phone, laptop, or AI system. It also does not explain how the components connected, how answers were generated, whether a third party participated, how many questions or answers were affected, or how investigators first detected the conduct.
The report does not quantify whether similar incidents have occurred in this examination or compare the alleged method with existing proctoring procedures. The allegations involving the separate LEET case remain allegations in this report. New Runtime has not independently verified the reported account, and the details may change as the investigation proceeds.
Skill to build now
If you operate an assessment or review process this week, map the full path by which a protected prompt, question, or document could leave the controlled environment and how an answer could return. Treat permitted hardware as an object to inspect for capabilities, not only as a category on an allowed-items list. For each path, decide which physical check, process constraint, or later review would detect or deter it, and document the privacy and fairness limits of that control.
Reader takeaway
The key lesson is not that AI alone defeats an exam, but that AI can amplify a physical information channel. For assessment builders, secure design has to consider how questions leave the room, how answers return, and whether permitted devices can be altered to do more than their ordinary purpose. The Han investigation is a reminder to examine the whole system, not only the screen.