What police say happened
Yonhap reports that the alleged ringleader is a self-taught hacker, aged 32, who began attacking organizations in September 2025. According to police, the group hit hospitals, golf courses, universities, private academies and subsidiaries of major conglomerates between September last year and October this year. The 8.2 million stolen records included personal information and internal corporate documents.
Some careful wording matters here. Sources describe the total differently. Some call it "records," and Yonhap says "pieces of personal data and other internal records." None of them says it means 8.2 million unique people.
The Herald Business gave a breakdown of the alleged victims:
| Victim category | Entities | Approx. records |
|---|---|---|
| Golf courses | 21 | 1.2 million |
| Private academies (hagwon) | 12 | 990,000 |
| Hospitals | 5 | 220,000 |
| Universities | 4 | 20,000 |
| Funeral service companies | 3 | 1.43 million |
| Insurance sales firms | 3 | 3.22 million |
| Other businesses | 42 | 1.12 million |
Decrypt's coverage also stresses that police were dealing with a mix of hospitals, schools and companies. The categories add up to 90 entities, while police say "more than 90." Treat the table as the outlet's reported breakdown, not a full census.
The Herald says the stolen personal data included names, resident registration numbers, phone numbers and addresses, and that no credit card data was taken. That is the outlet's account of police findings. It has not been independently confirmed for every victim.
The ARTEX angle
The tool at the center of the headlines is ARTEX. Reports say the suspect began with website vulnerability scanners. In late August he added an AI coding agent and an autonomous penetration-testing system. Police say he then breached 60 companies in 44 days, from August 31 to October 13. Local reports name the autonomous tool as Artex.
The Herald Business said that in the early stages, the ringleader manually operated automated tools that scan websites for weaknesses. After he incorporated generative AI, the scope of attacks reportedly expanded indiscriminately.
There is also a related but separate campaign. Police told Yonhap the tool is the same autonomous framework used against several South Korean banks, which was attributed to a Chinese-speaking threat actor. Yonhap's October 8 report on that campaign said CrowdStrike had described ARTEX as an open-source, China-developed AI penetration-testing tool. CrowdStrike assessed with moderate confidence that the bank attacker was likely a Chinese speaker and financially motivated. It did not name an adversary.
Sharing a tool does not mean sharing operators. Nothing in the reporting links the seven suspects to the bank attacker. ARTEX is publicly available, and that is the more worrying point. It also means the two cases may have nothing else in common.
What the reporting does not tell us
No source says what vulnerabilities were exploited, how the group got initial access, or what ARTEX did in each intrusion. There are no indicators of compromise and no Windows-specific details. If you manage Windows servers, IIS sites or SQL Server back ends, you can't conclude from this case that your stack was the target.
The "most of the attacks" claim, and the late-August start date for AI use, are attributed reporting. The English Yonhap piece supports AI use against 60 targets but does not name ARTEX. Treat the precise ARTEX share of the 90 breaches as unverified.
The background of the suspect is also described in different ways. One report says he studied arts and sports and has no IT background. The same report says police pointed out that he worked at a security company conducting simulated attacks from 2019 to 2022. The Herald Business says he learned techniques through blogs and YouTube and worked at a security firm for three years from 2019, simulating hacks on websites. "Self-taught" and "former pentester" can both be true. Professional experience in testing websites is a real head start.
Extortion: demanded versus collected
The extortion figures are easy to misread. Emails to 38 victims demanded a combined 360 million won, about $272,000. That is the amount demanded, not the amount paid.
SBS reports that only one company actually gave in, handing over just 300,000 won. The Herald Business puts actual losses at the same figure, about $227, and credits a swift police response. Some headlines suggest ransoms in the hundreds of thousands of dollars were paid, and they appear to be wrong.
SBS reports the group emailed victims samples of the hacked data, such as names, addresses and phone numbers. It warned that publicity would bring reputational damage and heavy fines. The Herald says that when victims asked to negotiate a lower amount, the gang threatened to leak the negotiation as well. SBS adds that the demands were for cryptocurrency. The group sent calls, texts and emails with short deadlines. That is a pressure tactic aimed at rattling staff before they can check facts or involve legal and security teams.
How police caught them
The investigation was ordinary police work with some AI help.
- A victim company filed a complaint in June.
- As similar reports came in from across the country, the Seoul Metropolitan Police Agency's Cyber Investigation Division took the lead in August, per the Herald Business.
- Between July and September, police arrested five accomplices who had phoned and texted victim employees.
- In early October they caught a man in his 20s who allegedly directed the others over Telegram.
- Police then analyzed victims' access logs and related evidence to identify the ringleader. They arrested him on October 15 and detained him two days later.
BleepingComputer gives October 23 as the date the ringleader was referred to prosecutors. Yonhap's Korean report also gives October 23. Only the ringleader is in custody, and the other six were referred to prosecutors without detention.
The investigation used server logs from breached entities. That is a lesson for any IT team. If your logs are thin, short-lived or missing, you may not be able to scope your own breach, and police may not be able to help.
Korea JoongAng Daily reports police said they recovered all stolen records containing personal information. Yonhap's Korean report says police are still investigating whether the data reached the dark web. Recovery of files does not prove every copy was deleted.
Police also said they built an internal-network AI analysis system on a high-performance AI computer supplied by the National Police Agency. The Herald says they developed three AI-based investigative tools. These tools were used for evidence analysis and are not part of the attackers' toolkit.
Why this matters to IT admins
Police called this "the first case of arrest of a suspect in Korea" for large-scale AI-assisted hacking and data theft. Korea JoongAng Daily says he reportedly began using generative AI extensively in August, which significantly expanded the scale of the hacking operation. That claim is the police's own characterization and applies to Korea. It is not a worldwide first.
The wider lesson is not that AI invented a new kind of attack. The police framing, as reported, is that AI lowers the barrier to entry. The skills needed to run a campaign like this are lower than they used to be. That was already true of commodity scanners and now applies to more steps in an intrusion.
The sources give no controls that police prescribed. The following are general industry practice, not findings from this case:
- Patch and scan your own web-facing apps. Autonomous tools probe a wide range of sites cheaply, so internet-facing sites such as marketing, booking and membership pages are plausible targets.
- Keep logs long enough to investigate. Access logs from victims helped identify the ringleader here.
- Minimize stored personal data. Resident registration numbers, contact details and records held for years increase the harm and the leverage for extortion.
- Prepare an extortion playbook. Decide in advance who verifies a claim, who talks to legal and regulators, and who contacts police. Short deadlines are designed to defeat that process.
- Treat leak threats as incidents. A message with sample data suggests you may already have been breached.
- Test yourself. If a commodity AI pentesting framework can be pointed at your public IP range, you want to find the holes first.
Bottom line
Seoul police say a small crew, led by a man with some professional testing experience, used AI tools to scale up website attacks against about 90 organizations and extort 38 of them. Actual payment was tiny, but the exposure of personal data is not.
Key caveats remain. The method of intrusion is not public, ARTEX's exact role is attributed reporting, and the link to the bank breaches is limited to tool overlap. Investigators are still looking at whether any of the data surfaced on the dark web.