One man with no IT degree, an AI coding agent and an autonomous penetration-testing system took 8.2 million personal records from 90 South Korean organizations in thirteen months, then sent 38 of them an invoice for keeping quiet. The patients, parents and mourners whose names and children’s details were in that haul got no warning and no vote — and the security industry is where he learned the trade.
The Seoul Metropolitan Police Agency’s cyber investigation division put the case on the table on Monday. Seven people have been detained or are under investigation. The alleged ringleader is 32, was referred to prosecutors on Friday while in custody, and has been identified in reporting only as Mr A. Six accomplices, some of them recruited off Telegram, are still being questioned. Yonhap’s account of the briefing puts the target list at more than 90 entities and the haul at 8.2 million pieces of personal data and internal documents.
Look at what the 90 were. Hospitals. Private academies — the after-school cram schools that most of Korea’s children pass through. Golf courses. Universities. A funeral services provider that lost 1.22 million records on its own. An affiliate of one of the country’s large conglomerates, and a regional mutual finance institution. Nobody on that list is a defence contractor. They are the soft middle of an economy: the places that hold ordinary people’s paperwork and treat a website login as a security policy. JoongAng Daily has the full shape of the target list.
The money is the part that makes you laugh before it makes you angry. Of the 90, 38 received threats. The group demanded a combined 360 million won — about $272,000. What actually landed in their accounts, according to Seoul Economic Daily, was 300,000 won taken from a single victim. About $210. That is the clearing price for eight million people’s private lives, and the crime turns out to be barely worth committing. The harm does not shrink to match the invoice.
Oh Gyu-sik — The Number That Should Scare You Is 40, Not 8.2 Million
Oh Gyu-sik, who heads the cyber crime investigation unit that briefed reporters on Monday, handed out a timeline more alarming than the body count. The group operated for about thirteen months, from September last year until this month. For most of that stretch the ringleader was doing what a competent amateur does: running vulnerability scanners against websites by hand, finding holes, taking what sat behind them.
Then he leaned on generative AI, and the arithmetic changed. From August 31 to October 13 — a little over six weeks — his operation broke into 60 companies using an AI coding agent and an autonomous penetration-testing system. Sixty in forty-odd days, after a year of doing it the slow way. The Seoul Economic Daily reconstruction is the one to read here, because it separates the two phases and shows exactly where the curve bends.
Thirteen months to learn a trade. Six weeks to industrialise it. That is not a story about one clever criminal. It is a story about a floor being removed from under every organisation that assumed the difficulty of attacking them was their protection.
The AI Agent — He Did Not Get Better at Hacking. He Got Something Else to Hack for Him.
Read the description of the tools carefully, because this is the whole argument. An AI coding agent and an autonomous penetration-testing system. Not a chatbot answering questions about injection flaws — an agent doing the enumeration, the probing and the exploitation work that used to demand either years of practice or a hireling who had them. Police suspect AI tools were used against 60 of the 90 targets, as The Korea Times reported.
What AI-mekna removes is not the knowledge. It is the labour. The gap between “knows roughly how this is done” and “can do it to 60 organisations in six weeks” used to be the gap that kept small-time criminals small. That gap is now a subscription and a long weekend.
Note what the police found alongside it: the same autonomous testing tool turned up in a recent breach across Korea’s financial sector. Investigators have not established a link between the two cases, and they were careful to say so. They did not have to. The tool is not bespoke. It is available, which is the entire point.
A Funeral Services Provider — 1.22 Million Grieving Families Were the Single Biggest Haul
The largest single loss: 1.22 million records, from a funeral services company. Think about who is in that database. Not customers in the commercial sense — families in the worst week of their lives, who handed over names, addresses and relationship details in order to arrange a burial. Fifteen percent of the entire haul came from one business that exists to serve people on their worst day.
Police say the stolen records have all been recovered. That cannot undo the months they sat in the wrong hands.
Korea’s Cram Schools — Minors’ Data, Stored as an Asset, Defended as an Afterthought
The private academies are the other half of the human cost. Their stolen files included personal information on children — students whose parents enrolled them in evening classes and handed over the household’s details to do it. A cram school is a small business with a sign-up sheet. It is not staffed to fight an AI-driven intrusion at four in the morning, and it should not have to be.
Which is the uncomfortable question the arrest does not answer. If 90 organisations across hospitals, universities and finance were all running the same class of unpatched website, then this was never 90 separate failures of diligence. It was one systemic failure with 90 addresses, and putting a 32-year-old in handcuffs does nothing about the other 89 doors.
The Telegram Recruits — Seven Dollars a Call to Become an Accessory to Extortion
The collection side of the operation was staffed the way a delivery app is staffed. Adverts went up on Telegram for people to make simple calls for pay; one promised a 30-second call was all it took. Recruits were paid 10,000 to 30,000 won per call — seven to twenty-one dollars — to ring executives and staff at the breached companies and remind them, in a human voice, that the data was out and the regulatory fine and the customer lawsuits were coming.
A 24-year-old handled accounts and recruitment. The rest ranged from their twenties to their fifties. Not one of them needed to understand a line of code. They needed a phone. The last mile of an extortion racket is now gig work, priced per call, and that is a labour-market fact as much as a crime statistic.
The Security Firm — It Trained Him to Simulate This Attack, Then AI Made the Simulation Real
Here is the detail that should end any remaining faith in the talent pipeline. The ringleader majored in arts and physical education. He had no IT background, and told investigators he picked up hacking from YouTube and blogs. But he also spent 2019 to 2022 at a security company, where his job was running simulated attacks — the mock hacking firms do to find their own holes before somebody else does.
So the industry spent three years teaching a man precisely how its clients’ websites fail, then returned him to the world with that knowledge and no leverage over what he did next. It is the oldest story in security: the person who tests the lock is the person who knows which one is fake. What is new is that the employer’s hold on that knowledge expired the moment a model could execute on it.
The containment systems worked right up until the person they trained stopped needing the job.
Your Own Records — Worth $210 to Him, and Worth a Decade of Trouble to You
Run the two numbers together and you get the actual state of play. On one side: 8.2 million records, a thirteen-month campaign, an AI stack, a phone bank of Telegram contractors, and a payoff of about two hundred dollars. On the other: a hospital patient, a cram-school parent, a family that buried somebody last winter — all now holding data that will resurface in somebody’s phishing campaign in 2029.
The economics have collapsed and the harm has not, which leaves enforcement as the only lever. Enforcement moved fast here only because it had help. Police ran their own AI analysis system over the evidence and logs and had all seven in custody within four months of a complaint filed in June. Korea’s first large-scale AI hacking case was closed by an AI reading the receipts.
That is where this goes. Attackers keep getting cheaper. Defenders keep getting automated. The 8.2 million people in the middle keep finding out by email, late, from a company they trusted with a form.
Sources: Yonhap, Korea JoongAng Daily, Seoul Economic Daily, The Korea Times.