South Korean Companies: AI-Assisted Hacking Gang Arrested Over 8.2M-Record Extortion Campaign

South Korean police have arrested seven people, including a ringleader in his 30s identified only by the surname A. Police say the gang used generative AI to break into the systems of about 90 domestic companies, take more than 8.2 million personal information records, and demand about 360 million won (roughly US$250,000) by threatening to leak the data. Details of the arrests come from SBS, which describes this as the first time South Korea has caught suspects in a large-scale hacking case that used generative AI. SBS also reports that the gang used the same AI autonomous penetration system seen in last month's breaches of South Korean financial institutions. Vendor research from CrowdStrike and Zscaler identifies that tool as ARTEX, an open-source agentic pentesting framework developed in China. None of the available sources says the arrested gang carried out the bank breaches. The two cases share a tool. They have not been shown to share an operator.

What Happened

SBS reports the following about the case:

• Timeline: The gang was active for about a year, starting in September 2025.
• Victims: About 90 companies, mainly private academies (hagwons), hospitals and golf courses. These are mid-sized organisations, not banks.
• Extortion: The gang told victims that a public leak would damage their reputation and bring administrative fines from regulators. It demanded about 360 million won in total. Only one company paid, and it handed over just 300,000 won (about US$210).
• Ringleader: Suspect A does not have an IT background. He reportedly taught himself hacking from blogs and other online sources.
• Investigation: Police say they built an internal AI analysis tool to track the gang. Seven people are in custody, and investigators are looking at further charges.

Every one of these details comes from a single broadcast report (SBS, tagged OTHER), which attributes them to police. None of the PRIMARY sources available for this brief covers the arrests. The victim count, record total and extortion figures should be treated as police claims reported by the press until a formal police statement or indictment is published.

Context: the financial-sector wave. The "recent financial sector personal information leak" SBS mentions is the campaign that hit South Korean banks from late September 2026:

• Victims named: The FSC confirmed a breach at Shinhan Bank and incidents at KB Kookmin Bank (BleepingComputer, Reuters via NBC). Yonhap reported that Hana Bank and Woori Bank were also breached. DW reports that seven major financial institutions were affected in total.
• Wider scope: The Korea Financial Security Institute said two savings banks were under investigation for the same type of attack (Herald Business).
• Response: The FSC held an emergency meeting, and President Lee Jae Myung ordered a full investigation.
• Other sectors: According to DW, Korea Electric Power Corp. and two large churches later confirmed unauthorised access. DW notes it is unclear whether the same perpetrator was behind those intrusions.

Accounts differ on who is behind what. CrowdStrike and Zscaler both assess, with moderate confidence, that the bank campaign was run by a financially motivated, Chinese-speaking actor. BleepingComputer, citing CrowdStrike, reports that a résumé the actor apparently generated with AI tools points to a 26-year-old living in Maoming, Guangdong. CrowdStrike itself says those personal details are not reliable enough to confirm an identity. The gang arrested in Korea has a different profile:

• Duration: It operated for a year, while the bank campaign lasted a few weeks.
• Targets: It hit small and mid-sized businesses, not banks.
• Location: Its members were arrested domestically.

The sources do not establish whether the two groups are linked.

What Was Taken

The figures below describe different incidents and should not be added together:

SBS does not break down which fields were in the 8.2 million records. Given the victim types, the data is likely to include student and parent contact details from academies, patient information from hospitals, and member records from golf clubs. Hospital data in particular can carry health information, which South Korea's Personal Information Protection Act treats as sensitive. This is an inference from the victim profile. No source has confirmed the specific data types.

For the bank campaign, Zscaler counts about 67,000 affected individuals across institutions. The per-bank figures in local media are lower than that, so the full total is still unsettled. CrowdStrike says the number of affected organisations "remains unconfirmed."

Why It Matters

AI lowers the skill barrier. By the police's account, a self-taught operator with no IT background compromised about 90 organisations using an off-the-shelf agentic pentesting framework. This is what defenders have been warning about: the AI handles reconnaissance and exploitation, and a low-skill operator runs it at scale. The Financial Security Institute adds an important qualification from the bank investigation: "the AI did not act independently without human involvement. A hacker used the AI as a tool" (Herald Business).

One tool, more than one operator. ARTEX was openly available on GitHub. If SBS is right that the arrested gang used the same system as the bank intruders, then at least two unrelated operations had the same capability. Defenders should treat ARTEX-style tooling as a commodity and should not expect it to be tied to a single actor. Arresting one group does not take the tool off the market.

Small and mid-sized organisations are the soft target. Academies, clinics and leisure businesses hold large volumes of personal data but rarely have dedicated security teams. The gang leaned on the threat of administrative fines from regulators to pressure them. That tactic only works where victims see a regulator as a bigger risk than the attacker.

The extortion largely failed. About 90 victims produced one payment of 300,000 won. That suggests Korean businesses mostly refused to pay. It also means the gang's main profit may have come from selling the data rather than from ransoms, so those 8.2 million records may already be in circulation.

Investigators are adopting AI too. Korean police say they built their own AI analysis tool to trace the gang. The FSS and the Financial Security Institute shared 28 attacker IP addresses across the financial sector (Reuters via NBC). That sharing is how the institute found additional bank victims (Herald Business).

The Attack Technique

SBS reports only that the gang used generative AI and "the AI autonomous penetration system" from the financial-sector breaches. Its specific intrusion paths have not been disclosed. What is known about that tooling comes from the PRIMARY vendor analysis of the bank campaign:

• Tooling: ARTEX is an open-source, LLM-driven agentic pentesting framework built in China and aimed at Chinese-speaking users. It automates reconnaissance and exploitation (CrowdStrike, Zscaler, Herald Business).
• Model backends: CrowdStrike found that the bank actor's ARTEX instance ran on DeepSeek v4.1-flash, probably reached through the LLM API reseller xcai[.]pro. The actor added GLM-5.3 (Zhipu AI) and Grok 4.6 in separate Claude Code sessions (BleepingComputer, quoting CrowdStrike).
• Initial access: The attacker exploited peripheral internet-facing web applications rather than core banking systems (Zscaler). Examples include a loan-progress inquiry service for brokers at one bank, an employee mobile work-support system at another (CrowdStrike), and Hana Bank's sales-support system (BleepingComputer).
• Infrastructure: CrowdStrike tied 38.244.50[.]120 to an ARTEX deployment. The actor's open directories exposed Claude Code session histories, ARTEX configuration files and Claude memory files. Yonhap reported that a server used in the attacks had an HTML page title containing a Chinese-language ARTEX string.
• IP rotation: The FSS identified 28 IP addresses in the United States, Japan, Germany and at least ten other countries (DW). A Financial Security Institute official said attackers "keep switching IPs, but the method is identical" (Herald Business).

Because the gang's targets were less defended than banks, it is plausible that it used the same approach of automated discovery and exploitation of exposed web apps and weak authentication. That has not been confirmed.

What Organizations Should Do

1. Inventory every externally reachable system, not just customer-facing ones. The FSC told financial firms to inspect all internet-exposed services, including staff portals, broker tools and mobile app backends (BleepingComputer). The bank intrusions began on exactly these overlooked surfaces.
2. Fix authentication and access controls on peripheral apps. Look for unauthenticated API endpoints, insecure direct object references, and inquiry portals that return more data than the user needs. Put employee and partner apps behind zero-trust or VPN-gated access instead of leaving them on the open internet.
3. Hunt for agentic scanning patterns. AI-driven frameworks produce fast, methodical, iterative probing from IP addresses that keep changing. Rate-limit and alert on bursts of enumeration against web apps. Block or investigate published indicators such as 38.244.50[.]120 and the 28 IPs the FSS shared with the sector.
4. Minimise and segment personal data. Academies, clinics and membership businesses should purge stale records, encrypt sensitive fields such as health and ID data, and keep personal data stores separate from web-facing application servers.
5. Prepare an extortion response before you need one. Report incidents promptly to KISA and the Personal Information Protection Commission. Doing so removes the attacker's main lever, the threat of regulatory fines. Do not pay. The near-total refusal by this gang's victims shows that refusing works.
6. Seed deception. Canary credentials, decoy endpoints and honeytokens in exposed apps are effective against automated tools that follow every lead. Zscaler recommends deception controls specifically against AI-assisted intrusions.

Sources: Gang Arrested for Stealing 8.2 Million Personal Info Records Using... | Unknown Threat Actor Uses AI-Driven ARTEX to Target South Korean Fi... | Threat Actors Abuse AI Pentesting Framework ThreatLabz | Hacker used ARTEX AI and Claude agents to target South Korean banks | South Korea probes bank breaches amid suspected AI-powered attacks | Exclusive: Chinese AI tool ARTEX used in wave of bank hacks, probe... | South Korea’s Lee says AI appears to have been used in bank hacks | S. Korea warns of AI-aided hacking after bank data breaches