In November 2026, news broke that the top scorer on Korea’s Certified Tax Accountant exam (세무사 시험) had personally admitted, on their lawyer’s YouTube channel, to cheating with generative AI and electronic devices, and had turned in a written confession to the police.

It wasn’t smart glasses. By the person’s own account, they photographed the exam questions with a miniature camera, sent them to an AI, and checked the answers on a smartphone. They did not say how they managed to see those answers in front of the proctors. In online communities, people speculated that since looking at a phone screen in the exam hall would have been difficult, they must have listened through a tiny earpiece. That is guesswork, though, not the person’s own explanation.

As I took in the news, the gambling table from Tazza: The High Rollers (〈타짜〉), the Korean film about card sharps, suddenly came to mind.

Those Who Cheat and Those Who Catch Them

In Tazza, what decides the game is not always just the luck of the draw or skill.

Cards get swapped by sleight of hand, opponents’ eyes are deceived, and sometimes the whole game is rigged by design. One person watches the other’s fingertips to spot the trick, while someone else prepares a new technique to slip past that watch.

A scene from the film Tazza (2006) in which a character, holding cards in front of a pile of banknotes on the hwatu (Korean flower card) table, stares at an opponent

A scene from the film Tazza (2006). Cine21 (씨네21)

It is a world where the art of cheating and the art of catching cheats keep evolving without end.

A scene from the film Tazza (2006) in which four characters surround a table covered with bundles of banknotes and hwatu cards, sizing each other up

The film Tazza (2006). Cine21

What is happening in national exams today looks rather similar.

Ban smartphones from the exam hall, and smaller devices appear. Check for smart glasses, and people turn to gear that isn’t eyewear. Crack down on visible electronics, and they pair up cameras and earpieces that are easy to hide on the body.

Close-up of a hand holding a pencil next to a hand holding a small slip of paper with formulas written on it

From crib notes in the palm to miniature devices. Wikimedia Commons (CC BY-SA 4.0)

With generative AI added to the mix, the nature of cheating has changed too.

In the past you needed an accomplice on the outside to feed you the answers; now AI can take over part of that role. That doesn’t mean it solves every question correctly, of course. But the cost and technical barriers to tapping outside knowledge in real time have dropped sharply.

The problem is that there is no end in sight to this race.

How Far Should the Exam Hall Be Sealed Off?

Let’s stay in the world of Tazza a little longer.

Imagine that, to prevent cheating at the gambling table, the players’ belongings and clothing are strictly controlled. What if they even had to sit at the table in nothing but their underwear, so they couldn’t hide anything on their bodies?

In fact, the sequel Tazza: The Hidden Card (〈타짜: 신의 손〉, 2014) has a scene in which the characters sit at the hwatu (화투) table, playing Korean flower cards in only their underwear. It’s a scenario that is hard to laugh off as a mere joke.

A film scene in which three shirtless men sit around a round table in a dark room with hwatu cards and piles of banknotes in front of them

A scene from the film Tazza: The Hidden Card (2014).

What happens if we apply that logic to the exam hall?

Glasses are inspected, ears are checked, radio signals are blocked, and cameras are installed throughout the hall. Even after the exam ends, the footage is analyzed for suspicious behavior.

Technically, AI really can help with part of this process.

Computer vision could detect suspicious electronic devices, or pick out particular behavior patterns in the footage and ask a proctor to take a closer look.

One could also consider a setup in which AI detects smart glasses or miniature electronics at the entrance, analyzes anomalous behavior during the exam, and uses the records afterward to investigate suspected cheating.

An AI exam-proctoring dashboard showing each test-taker's webcam and screen alongside a log of violation alerts

Example of an AI exam-proctoring screen (test-taker screens and alert log). TestInvite

Of course, technical feasibility does not by itself justify wholesale adoption.

An AI detection result is not, in itself, proof of cheating. False positives could leave innocent test-takers wrongly accused, and we also have to ask whether filming and analyzing every examinee is justified. The storage and use of video records, privacy protection, and appeal procedures are important issues as well.

Still, we can’t simply leave new forms of cheating unchecked.

Using AI to counter AI-assisted cheating. That is a realistic response well worth considering.

But the technology’s accuracy, privacy protection, a human’s final judgment, and an appeals process all have to be designed in together.

And yet, will the race end there?

Let’s imagine something a bit more extreme.

Neuralink, co-founded by Elon Musk, is developing a brain-computer interface (BCI) that connects the human brain directly to a computer. At present it is in clinical trials of technology that helps people with physical impairments operate a computer by thought alone.

A composite image showing a small implant inside the head of a person in profile, with the Neuralink logo on the right

A composite image depicting Neuralink’s brain implant

It is still a long way from delivering exam answers to a human brain or injecting knowledge into it.

But what if we follow the direction of the technology a little further out?

Suppose, for instance, that at some point in the future a small chip implanted in the brain let you exchange information with an external AI in real time.

Now picture a scene straight out of a movie.

There is a wealthy parent in Gangnam (강남), Seoul’s affluent district known for its education fever. Getting their child into medical school has been the goal for years.

In the private education market, paying enormous sums for the sake of college admissions is already nothing unusual. The misuse of prescription drugs to boost concentration for exams has also been a matter of public controversy. Ultra-expensive admissions consulting exists as well.

But what if, one day, a technology appeared that let you pay tens of millions or even hundreds of millions of won to get help from an AI connected to your brain for just one day, the day of the exam?

Let’s assume conditions so favorable they’re unrealistic: the surgical risk has become low enough, and everything can even be restored to its original state once the exam is over.

Can we say with confidence that not a single parent would make that choice?

Students studying with workbooks open at a study cafe near the Daechi-dong cram school district at around 10:30 p.m.

A study cafe at 10:30 p.m. in Daechi-dong (대치동), Seoul’s cram-school hub. Herald Business (헤럴드경제)

Of course, this is not something today’s Neuralink technology can do. A real brain implant is no minor procedure, and using one for a single day and then removing it is not a realistic technology either.

But the question this thought experiment raises is not about when the technology arrives.

Even if we check for smart glasses, find earpieces hidden in ears, block radio signals, and install dozens of cameras in the exam hall, what will we do if the means of cheating someday moves inside the body?

Will we then have to scan the brains of examinees as they walk into the hall?

Just as the players’ bodies and belongings are tightly controlled to stop cheating at the gambling table in Tazza, won’t the exam halls of the future come to demand ever more extreme surveillance?

There is another paradox here.

The more social cost we pour into preventing cheating, the more we may lose the room to ask what the exam is actually measuring.

A five-step staircase diagram rising from a bag check, a handheld detector, a signal-blocking shield, and a CCTV lens to the outline of a head with a glowing chip

The surveillance escalation ladder (diagram drawn by the author).

That is not to say we should let cheating slide.

On the contrary, we should actively consider AI-based detection technology to respond to new forms of AI-assisted cheating.

But at the same time, there is one question we must not lose sight of.

Why do we have to pay so much to protect the result of a single exam?

And does that exam really do enough to prove a person’s future capabilities and expertise?

A Society Where Even the Genuinely Skilled Are Suspected

There is one more cost to this problem.

Failing to catch someone who actually cheated is a problem, but suspecting someone who legitimately earned a high score of being a cheater is a serious problem too.

An overwhelming score on a single exam used to be something to celebrate.

But the more widely AI-assisted cheating becomes known, the more an outstanding score can itself become an object of suspicion.

Close-up of a hand holding a computer-readable marker pen and marking answers on a CSAT OMR answer sheet

A hand marking answers on an answer sheet for the CSAT (수능), Korea’s college entrance exam. JoongAng Ilbo (중앙일보)

People who did not cheat end up having to offer extra proof of their innocence.

In economic terms, this is a problem of the exam score losing its reliability as a signal.

When trust in an exam’s fairness collapses, the damage doesn’t fall only on the cheaters. Examinees who studied honestly, consumers who trust the credential, and the institutions that run the exam all bear the cost.

So preventing and catching cheating in the here and now remains important.

But we need to go a step further and ask a question.

Why Do We Have to Guard the Result of a Single Exam This Fiercely?

National exams are not simple tests of knowledge.

Some determine who gets a chance at university; others grant the right to enter a profession. A single pass can shape decades of career opportunities and financial rewards.

That is how much exam fairness matters.

But in an era when AI is drastically lowering the cost of accessing knowledge, is the ability to solve problems without outside help, on a particular day in a particular place, still the most appropriate standard for judging expertise?

In actual work, the person who uses AI to solve problems faster and more accurately may well be the one who is rated highly.

Yet in the exam hall, the moment you use that same tool, it becomes cheating.

A screen where a generative AI writes a draft paragraph of a report after a one-line question is entered

Drafting with AI counts as skill at work but as a foul in the exam hall. Wikimedia Commons

Of course, breaking the established rules is cheating. That fact doesn’t change.

But whether those rules will go on properly measuring the abilities we want to measure is a separate question.

We need to advance the technology for stopping cheating with AI.

At the same time, we need to think about what we call ability in the age of AI, and how it should be verified.

That is because there are problems that can’t be solved just by controlling the exam hall ever more perfectly.

A story that began with the gambling table in Tazza ends up at a slightly different question.

We are focused on how to administer exams fairly. But are we asking enough about what those exams are supposed to prove?

An empty classroom after the examinees have left, with desks and chairs arranged in neat rows

An empty classroom after the exam. Wikimedia Commons (CC BY-SA 4.0)


Next post: “Cheating in the Exam Hall, Skill at the Office?” (「시험장에서는 부정행위, 회사에서는 실력?」) — what abilities should we be measuring in the age of AI?