Four of South Korea's largest banks disclose data breaches within days of each other
TLDR
- The FSC says 16 South Korean financial firms confirmed breaches by 14 October, exposing about 100,000 personal records.1416
- Investigators traced the attack tool to open-source penetration-testing system ARTEX AI, now closed-source, though derivatives circulate.713
- CrowdStrike assesses with moderate confidence that the attacker is a financially motivated Chinese speaker, not tied to a named group.11
Update, 17 October, what changed: Yeogaram Savings Bank has published its final tally: 43,246 customers had personal information leaked, below the roughly 50,000 Financial News says was reported earlier (earlier updates here carried "around 40,000 customers" and spell the bank "Yegaram").17 Of those, 42,926 (99.3 per cent) had basic details exposed, while 320 also had sensitive data such as resident registration numbers, account numbers, transaction histories or passport numbers.17 The bank said the attacker accessed its server on 16 September but it only became aware of the leak on 30 September, 14 days later, and that no secondary harm had been confirmed when it issued its notice on 16 October.17
Separately, Aju Press reported on 16 October that Threatbook, a Chinese security firm, has announced a cloud platform called RedAgents.ai that will host several AI agents, including a paid ARTEX PRO, the commercial version of the ARTEX open-source AI penetration-testing tool.18 Aju Press says the announcement came from Threatbook's chief executive, Xu Feng, that the ARTEX developer is a Threatbook threat-detection engineer who built the tool as a personal project, and that the developer had made the public GitHub repository private and suspended the project over misuse concerns.18 That account of the tool's origin differs from the earlier reporting in this article, which described it as built for a Baidu-run security challenge; the two accounts are as reported and have not been reconciled. Aju Press repeats that evidence indicated ARTEX was exploited in the Korean attacks but does not say Threatbook or its staff were involved in them.18
Update, 15 October, what changed: the Financial Services Commission said at a second emergency meeting on 14 October that 16 financial firms have now confirmed breaches, up from the seven first reported and from 11 the day before, with about 100,000 personal records exposed.1416 Five firms were newly confirmed: DB Savings Bank, KDB Capital, the electronic financial services provider Brightfix, and the cryptocurrency businesses Streami (operator of the Gopax exchange) and Prabang.141516 KDB Capital said it detected abnormal external access on 12 October and that names, resident registration numbers and dates of birth of 848 customers were leaked, Prabang reported contact details and addresses of 20 customers, and DB Savings Bank said an attempt was blocked with no customer data leaked.15 Gopax said attackers logged in using previously leaked IDs and passwords, that fewer than 10 customer accounts were accessed abnormally and that there was no asset damage.16 Regulators have found no evidence that account or card numbers were compromised.141516 They are investigating whether the same hacking group was behind all the attacks.14
The meeting also exposed wider gaps: of 555 financial firms whose security checklists were reviewed through 8 October, 130 (23.4 per cent) reported deficiencies, mainly missing security patches, unsupported ageing systems, no 24-hour monitoring and missing authentication for externally accessible systems, with electronic finance and online-investment-linked firms weakest.1415 The FSC plans to map all external connection points, clarify the responsibilities of chief executives and security officers, and raise vulnerability checks at large firms from once to at least twice a year.1415 Outlets spell the crypto exchange as Prabang (Seoul Economic Daily, ChosunBiz) or Pravang (Korea Times).141516
Update, 10 October, what changed: BleepingComputer reported that the attacker likely obtained access to the DeepSeek backend used by the ARTEX instance through a third-party API reseller, xcai[.]pro, rather than a direct DeepSeek account.13 Separately, ARTEX AI's developer took the project closed-source after confirming it had been used in real-world attacks, though English and Korean-language derivatives built from the tool's earlier open-source code remain available and in circulation.13
Update, 8 October, what changed: CrowdStrike Intelligence identified infrastructure tied to the campaign and assessed, with moderate confidence, that the attacker is likely a Chinese speaker acting for financial gain, based on the use of the Chinese-developed ARTEX tool and Chinese-language prompts found in the attacker's own files.11 The attacker has not been attributed to a named group.11
CrowdStrike found an open directory at a Hong Kong-based IP address containing Claude Code session histories, ARTEX configuration files and Claude memory files, revealing a two-server setup: the Hong Kong address as the attacker's primary infrastructure, and a second IP address hosting the ARTEX instance likely used against the Korean banks.11 The ARTEX instance ran on DeepSeek v4.1-flash as its main model, supplemented with GLM-5.3 and Grok 4.6 in separate Claude Code sessions, with DeepSeek access likely obtained through a proxy/reseller service.11 Beyond the pentesting activity, the attacker used Claude to ask where stolen Korean data is typically sold and to find Korean-language Telegram groups dealing in breached data.1112
In one session, the attacker asked Claude to draft a security-researcher résumé listing the ARTEX-related work as an achievement, and the prompt included personal details: the name "YY," a phone number, the Telegram handle @YY520CN, an age of 26, education at South China University of Technology, and a location in Maoming, Guangdong.11 CrowdStrike said the same Telegram handle also appeared in unrelated Claude Code sessions researching a Telegram-based NFT marketplace and in activity targeting a Chinese payment platform, but cautioned that currently available information cannot definitively tie those personal details to the ARTEX operator.11 CrowdStrike said compromised systems in the campaign it reviewed included a bank's loan-inquiry service used by brokers and a separate bank's employee mobile work-support system, consistent with earlier reporting, and that the attacker's identity, full breach scope and amount of stolen data remain unconfirmed.12
Update, 7 October, what changed: South Korea's National Police Agency launched a full-scale criminal investigation on 6 October, assigning 28 investigators across four teams from its cyberterrorism unit, after the Financial Supervisory Service separately identified 28 distinct IP addresses linked to the hacking attempts.9 President Lee Jae Myung called for a swift response at a Cabinet meeting the same day, saying "we should accelerate the development and deployment of artificial intelligence technologies specialized in cybersecurity. It is time to overhaul our society's security paradigm for the AI era," and warned that technological advances were outpacing defences.9 Police are also weighing whether the case must be referred to the newly launched Serious Crimes Investigation Agency, pending the Financial Services Commission's determination of whether the affected systems count as regulated electronic financial infrastructure.9
Separately, AhnLab's Security Emergency response Center (ASEC) traced the ARTEX AI tool's hosting infrastructure and found roughly 600 internet addresses worldwide running ARTEX instances, well beyond the servers already tied to the Korean bank breaches.10 ASEC also found a second open-source AI attack framework, CyberStrikeAI, which supports multi-agent orchestration, web-shell deployment and command-and-control management, co-located with ARTEX AI on some of the same servers, raising the possibility that multiple AI-driven attack tools are being used together in the wild.10 ASEC cautioned that because both tools are public and open-source, the presence of an instance does not by itself prove malicious use or a shared operator; some servers could belong to red teams, penetration testers or security researchers.10 Financial-security sources told Yonhap that the same attacker IPs recur across the breaches at Shinhan, Kookmin, Hana, BNK Busan, Yegaram Savings Bank and Hyundai Capital, and that ARTEX AI traces have now also been found in those attacker IPs, with the National Police Agency's cyber investigation bureau continuing its probe.10
Update, 6 October, what changed: the campaign has now hit seven financial institutions and exposed more than 65,000 customer records, up from the original four-bank cluster, triggering a round-the-clock cybersecurity emergency response from South Korean authorities.7 Investigators say attackers were undetected inside bank systems for as long as 43 hours before containment measures took effect.7
The Korea Financial Security Institute traced the attack tool to ARTEX AI, an open-source, agentic penetration-testing system that can independently run reconnaissance, find vulnerable login endpoints, launch attacks and verify results with little human direction.7 The institute identified it after finding the Chinese-language string "ARTEX-自主渗透测试控制台" ("autonomous penetration testing console") in the HTML title of a server used in the Shinhan Bank attack.7 Unlike the state-sponsored tooling that has previously targeted Korean financial infrastructure, ARTEX AI was not built by a military or intelligence programme; it was submitted as a winning entry in a challenge run by Baidu's Security Response Center and distributed openly on GitHub, mainly to Chinese-speaking developers, before a threat actor picked it up and pointed it at internet-facing banking portals.7
Official channels have not confirmed ARTEX AI's use in the breach, and the Chinese-language string does not by itself tie the campaign to a specific threat actor.8 South Korea's Financial Services Commission convened a separate emergency meeting after confirming the Shinhan Bank breach and incidents at Kookmin Bank, instructing financial firms to inspect every externally accessible system, including non-customer-facing ones, tighten authentication and access controls, and rapidly share threat intelligence.8 Reported victim counts vary by outlet: Korea Herald put Kookmin's leak at 119 customers' personal information, while BleepingComputer, citing local reports, put the figure at 119,000 customers' credit card information, a discrepancy regulators have not yet resolved.8
Update, 4 October, what changed: President Lee Jae-myung ordered a formal investigation and response measures after being briefed on the string of breaches, calling the matter serious; the order followed an emergency meeting Financial Services Commission chairman Lee Eog-weon convened on Sunday with industry associations, regulators and executives from affected institutions, originally scheduled for 7 October but brought forward after further breaches surfaced at second-tier financial institutions.5 The FSC chief said authorities could not rule out AI involvement in the attacks and called for an "AI attacks defended by AI" approach, alongside broader upgrades to the sector's cybersecurity framework.5
Regulators now believe the intrusions may have broadly scanned multiple financial companies for vulnerabilities rather than targeting a single institution, with attack traffic traced to IP addresses across the United States, Japan, Singapore, Vietnam and Britain, according to bank data submitted to lawmakers.5 Authorities directed financial institutions to run comprehensive security inspections, tighten access controls, minimise external system access, strengthen consumer protections and rapidly share attack methods and IP information across the industry.5 South Korea's main opposition People Power Party called for authorities to also investigate possible North Korean involvement, citing Pyongyang's history of cyberattacks against the country's financial institutions.6
Update, 3 October, what changed: security researchers confirmed the AI penetration-testing tool found on the server linked to Shinhan's breach is ARTEX AI, a Chinese-origin tool, moving the earlier "suspected" tool attribution to confirmed.3 The breach spread beyond the four largest banks into the secondary financial sector: Hyundai Capital confirmed on 2 October that an attack on a public mortgage-broker lookup page exposed personal data, including internal broker numbers and resident registration numbers, for 146 mortgage brokers, while Yegaram Savings Bank disclosed that a server breach detected on 30 September may have exposed names, dates of birth and contact details for around 40,000 customers, roughly a fifth of its customer base.4 Both companies said core customer banking-transaction data was not affected and reported the incidents to the Financial Supervisory Service and Korea Internet & Security Agency.4
Shinhan Bank disclosed on 1 October that an unauthorised party had bypassed identity verification in a mobile inquiry service used by loan agents to check application progress, exposing customers' names, phone numbers, annual income, calculated loan limits, 66 resident registration numbers and 97 CI identifiers used for online identity verification.2 The bank said it blocked the external IP addresses involved and suspended the affected service.2
KB Kookmin Bank said on 2 October that personal information on 119 customers, including names, phone numbers, addresses and encrypted resident registration numbers, was leaked after abnormal external access to a mobile system used by employees; it detected the possible breach on Wednesday night and blocked the affected server and access routes.1 Hana Bank separately reported that personal information on 89 customers, including resident registration numbers, names, addresses, email addresses, phone numbers and employer names, was exposed through its sales support system (ODS) after an external hacking attempt; the bank said the system is separate from its internet and mobile banking transaction platforms.2 BNK Financial Group reported that 11 records containing outsourced employees' personal information were leaked in a separate incident, while Woori Bank and NH NongHyup Bank said they faced hacking attempts that did not result in exposed data.2
All four banks said customer banking-transaction information was not affected and pledged to compensate customers for any resulting losses.1 The Financial Services Commission convened an emergency meeting with the Financial Supervisory Service, security officials and industry representatives, ordering checks of externally accessible systems and calling for stronger authentication, tighter access controls and faster threat-intelligence sharing, while vowing to investigate the attacks and draft regulatory improvements.1
Why it matters: a leaked AI-agent session log handing investigators a name, phone number, Telegram handle and hometown shows how attacker tradecraft built around chatty coding assistants can leave a far more personal trail than traditional malware ever did, even as the underlying attack tool, now closed-source, has already spun off open derivatives for the next operator to pick up.
Sources
- Shinhan, Kookmin, Hana data breaches fuel concerns over AI-powered cyberattacks in financial sector (The Korea Times)
- Hana, BNK report data breaches as hacking hits Korean banks (The Korea Herald)
- [단독] 은행 연쇄 해킹 도구는 중국 '아르텍스 AI'…해커가 사용했다 (Herald Business)
- AI 해킹에 비상 걸린 韓금융…은행 이어 현대캐피탈·예가람저축은행도 뚫렸다 (Newsis)
- South Korean president orders probe into wave of data breaches at banks, public agencies (The Business Times)
- Lee Orders Probe Into AI-Powered Cyberattacks as Data Breaches Hit Shinhan, Kookmin, Hana and BNK Banks (IBTimes KR)
- Open-Source AI Agent Hacked Seven South Korean Banks, Exposing 65,000 Records (Tech Times)
- South Korea probes bank breaches amid suspected AI-powered attacks (BleepingComputer)
- Police launch major probe as suspected AI hacks sweep through banks (The Korea Herald)
- 은행 해킹에 등장한 'ARTEX AI'…전세계 600개 IP서 확인 (Yonhap News Agency)
- Unknown Threat Actor Uses AI-Driven ARTEX to Target South Korean Finance (CrowdStrike)
- Chinese-speaking hacker possibly linked to AI-driven attacks on S. Korean banks: report (Yonhap News Agency)
- Hacker used ARTEX AI and Claude agents to target South Korean banks (BleepingComputer)
- AI-driven cyberattacks hit 16 financial firms, expose 100,000 records (The Korea Times)
- 16 Financial Firms, Including Crypto Exchange Gopax, Hit by Data Leaks in AI-Driven Hacking Wave (Seoul Economic Daily)
- Hacks hit 5 more Korea finance firms, lift breach total to 16 (ChosunBiz)
- Yeogaram Savings Bank: Personal information of 43,246 people leaked... “Became aware 14 days after hack” (Financial News)
- Korean Hacking Tool 'ARTEX' Reintroduced as Paid Service in China (Aju Press)