business 6 min read

Korea's Top Tax Accountant Cheated With AI — What It Means for Every

A near-perfect score on Korea's tax accountant exam led to an AI cheating admission — and an urgent reckoning for credentialing systems across East Asia.

  • South Korea
  • AI Cheating
  • Credentialing Reform
  • Professional Exams

When perfection looks like fraud

Han Won-seok scored 397.5 out of 400 on Korea’s first-stage tax accountant qualification exam. Three months later, he placed first overall with 663 out of 800 on the written portion.

That is not simply a good score. It is a score that, by any reasonable prior assumption, should have been extraordinary enough to raise eyebrows.

It did.

In South Korea’s intensely competitive professional-examination ecosystem, a near-perfect result on the first stage of the tax accountant exam is the kind of statistical anomaly that makes licensed practitioners sit up and wonder what happened. Within days of the results coming out, other accountants and tax lawyers were calling it “close to impossible.” Han had also posted his score publicly on an online community, explaining (falsely, as it turned out) that he had passed through self-study alone.

That posting was the catalyst. Had his score been merely very high rather than historic, the case might still be dormant.

On a Sunday evening video hosted by his lawyer Lee Don-ho, Han admitted the truth: he used a miniature camera to photograph exam questions, forwarded them to a generative AI system, and checked the answers against his smartphone during the test. He did not use the AI-enhanced smart glasses many had suspected. He declined to share further details of the method, saying he wanted to prevent others from copying it.

He also admitted to the same technique on the recent certified public accountant exam.

Four proctors saw nothing

The more alarming finding in this story is not that someone cheated, but that they got away with it until their own excellence made them a target.

Four proctors supervised Han during both stages of the examination. None reported suspicious behavior. None flagged anything. The Human Resources Development Service of Korea (HRD), which administers the exam, initially announced it found no concrete evidence of cheating based on those observations.

Cheating went undetected not because the security apparatus was sophisticated, but because the candidate behaved like someone who belonged there — and then scored in a range so high it fell outside anyone’s practical expectation of plausibility.

Once Han confessed, the agency apologized and said it would file a criminal complaint for obstruction of business. But the apology itself is revealing: it acknowledges a gap between what existing detection methods could catch and what a determined test-taker armed with consumer-grade technology can accomplish.

The economics of detection are breaking down

Here is the structural problem that Han’s case exposes:

Miniature cameras and smartphones cost hundreds of dollars, not millions. A capable language model is free or nearly free. The barrier to executing this scheme is trivial and falling every month. Meanwhile, credible anti-cheating infrastructure — thermal imaging, randomized item generation, one-on-one oral defenses, invigilation ratios that approach the cost of the exam itself — requires sustained investment.

There is a growing mismatch between the cost of cheating and the cost of preventing it.

East Asian credential markets already operate under extreme pressure. In South Korea, professional certifications are gateways to income, social standing, and family expectations. Han himself described anxiety over disappointing his parents and struggling to run his own business after leaving corporate employment. The personal motive matters less than what it signals: when the penalty for failure carries real social and economic weight, the incentive to find loopholes compounds.

What happens next for Korean exams

The Blue House responded swiftly. Policy chief Ha Joon-kyung chaired an emergency meeting on AI-aided cheating just days before the national College Scholastic Ability Test, known as the CSAT.

Ha urged the education ministry and other agencies to examine all possible risks and deploy tools including thermal imaging cameras. “The fairness of examinations is directly tied to public trust,” he said, adding that “no form of cheating should be tolerated.”

The CSAT is a different creature from the tax accountant exam — more standardized, longer administered, fewer open-response items. But the principle at stake is the same: if generative AI can be smuggled into a testing room and used to answer questions in real time, the credential that follows cannot claim to measure knowledge unless the architecture itself changes.

The reforms on the table fall into three buckets, each with its own trade-offs.

Hardware-based detection. Thermal cameras, metal detectors, signal jammers, and smart-glasses screening. These work at the perimeter but create friction, raise privacy concerns, and remain vulnerable to methods they were not designed to detect — such as the tiny camera-plus-smartphone setup Han used.

Test-design reform. Randomized item generation means no two candidates receive the same question set. Shorter, adaptive exams reduce the window for external assistance. Oral defenses replace or supplement written sections. Each of these moves raises administrative complexity and cost, and none eliminates risk entirely.

Behavioral and procedural layers. Stricter post-hoc audits, statistical outlier analysis on scoring patterns, mandatory disclosure of study methods, and heavier penalties for confirmed cheating. These are cheaper to implement but reactive rather than preventive.

Who wins, who loses, and what this means beyond Korea

Right now, the losers are clear. Candidates who studied honestly lose credibility by association. Employers and licensing bodies lose confidence in the signal that certification provides. Students facing the CSAT and similar exams lose faith that the playing field is level.

The winners are harder to name, but they exist. Test-design vendors and security-tech firms will see demand rise. Agencies willing to invest in adaptive testing and oral-defense protocols will produce credentials with stronger reputations. And candidates who genuinely earn their certifications will eventually benefit from a market where the badge means something again.

But the central insight is simpler and more unsettling: in any system where high-value credentials are scarce and access to general-purpose AI is ubiquitous, cheating will trend toward the cheapest available method until the cost of preventing it rises above the cost of the crime.

Han Won-seok did not orchestrate a sophisticated conspiracy. He used a camera, a phone, and a language model. That is the point. The technology required for this kind of fraud is no longer experimental. It is on every desk, in every pocket, and freely available online.

Korea’s response will matter beyond its borders. Japan, China, and Taiwan face the same credential pressures and the same AI availability. If Seoul chooses lighter-touch detection upgrades over structural reform, the loophole widens. If it invests in adaptive testing and behavioral verification, it sets a new benchmark.

The tax accountant exam was supposed to certify expertise. Instead, it certified how far a camera and a chatbot could carry you.