Police in Seoul have arrested seven people accused of using generative artificial intelligence to break into roughly 90 organizations and steal 8.2 million records—a case the Seoul Metropolitan Police Agency describes as the first time in South Korea that suspects in a large-scale AI-assisted hacking operation have been taken into custody. The agency announced the arrests on Monday.

According to police, the targets included hospitals, universities, private cram schools, golf courses and subsidiaries of large conglomerates—organizations that hold the names, contact details and records of ordinary people. One funeral services provider lost 1.22 million records, and one report said data taken from the private academies included information on minors.

The alleged ringleader is a 32-year-old who studied arts and physical education and has no IT background, according to local reports. He told investigators he taught himself hacking through YouTube and online blogs, though police say he also carried out simulated attack work at a security company from 2019 to 2022.

He began with website vulnerability scanners, software that probes sites for weak spots. In late August, reports say, he added an AI coding agent—a program that writes and runs code on its own—and an autonomous penetration-testing system, a tool that automates the mock break-ins companies normally hire experts to perform. In legitimate use, tools of that kind help organizations find and fix their own weak spots before someone else does, which is what makes their alleged repurposing notable. Local reports identify the autonomous tool as Artex. Police say he then breached 60 companies in 44 days, from August 31 to October 13.

Police said an operation of this scale once required an entire hacking group, and that AI now allows one person to pull it off in a short period. The case's own numbers underline the shift: 60 of the roughly 90 breaches came in those final 44 days of a campaign that otherwise ran about 13 months.

The campaign ran from September 2025 through October, and the investigation opened in June after one victim filed a complaint. Emails sent to 38 victims demanded a combined 360 million won, about $272,000, but the only confirmed payment was 300,000 won, roughly $210—a yield that is small both against the scale of the breaches and against the ransom demands reported in earlier AI-assisted cases overseas.

The case adds to a pattern of criminal AI use documented in recent years, mostly in reports from AI companies rather than police. The Seoul investigation breaks that mold: it comes from law enforcement, with arrests and a prosecution referral rather than a vendor's incident report. In 2023, WormGPT, a ChatGPT clone built without safety guardrails, was used to write convincing phishing emails—messages designed to trick people into handing over money or passwords.

In August 2025, Anthropic said a single criminal used its Claude Code tool to run a vibe hacking extortion campaign against at least 17 organizations, with ransom demands ranging from $75,000 to $500,000 in Bitcoin. In November, the company said a Chinese state-sponsored group used Claude to carry out 80% to 90% of an espionage campaign against about 30 targets, though only a small number of the intrusions succeeded.

Researchers at the universities of Cambridge, Edinburgh and Strathclyde pushed back on the hype in a recent study of 97,895 cybercrime forum threads. They found AI's main impact so far in already-automated areas such as SEO fraud, social media bots and romance scams, with little sign that coding tools helped low-skill actors beyond what pre-made scripts already allowed. The Seoul case is a police-documented counterpoint: investigators describe a suspect with no IT background whose breach count accelerated sharply once AI tools entered his workflow.

The ringleader was referred to prosecutors in custody on Friday and faces charges under South Korea's information and communications network law, as well as extortion and attempted extortion. Police said they will continue investigating the six accomplices and work to identify additional victims; 15 organizations have reported damage so far. The figures worth tracking from here are the gap between those 15 reports and the roughly 90 organizations breached, and the outcomes for the six accomplices still under investigation.