CrowdStrike links breaches at South Korean…

CrowdStrike Intelligence reports a campaign against South Korean financial organizations, active from late September to early October 2026, that resulted in exfiltrated data, and says open directories on actor-controlled servers held Claude Code session histories, ARTEX configuration files and Claude memory files. CrowdStrike assesses with moderate confidence that the unattributed actor is a financially motivated Chinese speaker, and says the ARTEX instance used DeepSeek v4.1-flash as its main model, with GLM-5.3 and Grok 4.6 used in further Claude Code sessions. Press reports name Shinhan Bank, KB Kookmin Bank and Hana Bank among the victims; press reports at least seven institutions and about 68,000 people, while CrowdStrike calls the number of affected organizations unconfirmed.

Why it matters

The actor's own agent session histories and tool configurations give direct evidence of how an open-source agent framework and a coding agent were likely used in the breaches.

Key facts

As stated in the sources, with where to find them.

CrowdStrike says the campaign ran from late September to early October 2026 and resulted in exfiltrated data. It has not attributed the activity to a named adversary and assesses with moderate confidence, based on the Chinese-developed ARTEX tool and Chinese-language prompts, that the actor is likely a Chinese speaker and financially motivated.CrowdStrike post, introduction
CrowdStrike describes ARTEX as an open-source agentic penetration-testing tool developed in China. BleepingComputer reports that after its use in real attacks was confirmed the developer made the project closed-source and stopped updates, while derivatives built from the existing code remain available. Fortune calls ARTEX a Chinese state-developed tool built on open models, in a passage citing Reuters for the actor's likely location; CrowdStrike does not describe it as state-developed.CrowdStrike post, introduction; BleepingComputer article; Fortune article
CrowdStrike reports that an actor-controlled server hosting an ARTEX instance exposed a Claude Code project instruction file holding a Chinese-language penetration-testing prompt, and that a second, Hong Kong-based server held Claude Code session histories, ARTEX configuration files and Claude memory files. It says these show extensive activity against South Korean financial organizations that overlap with those in industry reporting.CrowdStrike post, ARTEX Infrastructure Analysis
Per CrowdStrike, the ARTEX instance used DeepSeek v4.1-flash as its primary model, likely reached through an LLM API reseller, and the actor used GLM-5.3 (Zhipu AI) and Grok 4.6 for additional Claude Code sessions. CrowdStrike also says the actor asked Claude where Korean breach data is typically sold and for help finding Korean Telegram data-sales groups. The post does not say which backend model served each Claude Code session.CrowdStrike post, ARTEX Infrastructure Analysis
CrowdStrike reports that in one session the user asked for a résumé describing the results of the ARTEX activity and supplied personal details, but says available information cannot definitively tie those details to the actor. This record does not repeat them.CrowdStrike post, Possible Threat Actor Details
CrowdStrike says the number of affected organizations is unconfirmed as of its 2026-10-07 post. The Record (2026-10-06) reports breaches of at least seven financial institutions and personal data of at least 68,000 people reportedly exposed, with Shinhan Bank reportedly saying about 25,000 customers were affected. Fortune (2026-10-21) attributes an estimate of 68,000 people to the Wall Street Journal.CrowdStrike post, Details; The Record article; Fortune article
The Record reports, citing local news, that financial authorities found 33 IP addresses used in the attacks, linked by the Financial Supervisory Service to at least 12 countries, that the incidents came to light on 2026-09-30, and that the National Office of Investigation formed a team of 28 investigators. It reports President Lee Jae Myung saying signs had emerged that AI agents were used in at least some of the attacks.The Record article
BleepingComputer names Shinhan Bank, KB Kookmin Bank and Hana Bank and reports exposure of customers' personal data and credit card information and outages at some banks; The Record, citing the Korea Herald, reports exposed borrowing history, incomes, phone numbers and names. CrowdStrike, citing industry reports, describes a breached loan-inquiry service for brokers at one bank and a compromised employee mobile work-support system at another.BleepingComputer article; The Record article; CrowdStrike post, Details
CrowdStrike's evidence is the actor's tooling, prompts and session records; its post does not measure how much of each intrusion the agents carried out on their own. BleepingComputer describes the link to the bank breaches as high-confidence, a level CrowdStrike's post does not state.CrowdStrike post, Assessment; BleepingComputer article

Findings that cite this record

Aug 27, 2025

Attackers have used AI agents to carry out much of the work of real intrusions, from reconnaissance to credential theft and extortion, with people directing them at a few decision points.

Corroboratedobserved13 evidence records1 linked finding

Oct 22, 2026

Attackers have run real intrusions with publicly available, open-source agent frameworks such as ARTEX, PentAGI and Hermes Agent, not only with commercial AI products.

Corroboratedobserved7 evidence records

Key questions this bears on

How are attackers using AI agents in real operations?To run parts of intrusions at scale: vendors report a state-aligned actor and criminal groups using open-source agent frameworks and commercial models.