The earliest known attack date is Sep 27. The first report to the authorities came on Sep 30 (Shinhan). Disclosures followed from Oct 1 to Oct 4. The attack surface was mainly external-facing edge applications, and the methods included SQL injection and authentication bypass. A Financial Security Institute (FSI) official said it was not a fully autonomous attack without human involvement.
Confirmed victims
- Shinhan Bank
- KB Kookmin Bank
- Hana Bank
- BNK Busan Bank
- Yegaram Savings Bank
- Hyundai Capital
- Welcome Savings Bank
Targets
- Woori Bank (says it blocked the attack)
- NH NongHyup Bank (says it blocked the attack)
Attribution
- ARTEX basis
- None of the three Financial Services Commission (FSC) press releases of Oct 4, Oct 14 and Nov 5 mentions ARTEX. The ARTEX claim comes from an anonymous Financial Security Institute (FSI) official (Herald Business, Oct 3, scope limited to Shinhan) and the CrowdStrike report of Oct 7. CrowdStrike did not name the victim institutions.
- Actor
- CrowdStrike: financially motivated, likely Chinese-speaking (medium confidence), not attributed to a known group. Recorded Future tracks it as TAG-217.
- Same attacker?
- Yonhap reported on Oct 4, citing the authorities, that the same attacker IP was seen at the 7 institutions. The Financial Services Commission (FSC) changed this on Oct 14 to "under close review", and there has been no official confirmation since.
Figures and definitions
| Victim count | About 66,000 people, about 2,200 corporate records (Genians, as of Oct 4) |
|---|---|
| Victim count | About 67,000 people (Zscaler, Oct 14) |
| Victim count | At least 68,000 people (Recorded Future) |
| Victim count | About 68,000 (Weekly Kyunghyang, Oct 12, 7 institutions) |
| Shinhan | 25,729 (Herald Business, National Assembly audit materials); 25,727 (CEO Score and others) |
| KB | 119 people (Oct 2); corrected to 2,210 people (Oct 26) |
| Yegaram | About 40,000 (Oct 2); about 50,000 (Oct 13) |
| Attack IPs | FSS: 19 IPs, 12 countries (Financial News, Oct 6); 28 IPs, 12 countries (industry sources); 33, or 28 after deduplication (Genians); more than 20 IPs across more than 10 countries (WSJ, as relayed); CrowdStrike lists 9 proxy IPs |
| Reason for the block (Woori) | Five accounts: intrusion prevention system; no such vulnerability; multi-factor authentication; luck; Xint inspection (sources include the vendor) |
Related IoCs
| Type | Value | Role | Check |
|---|---|---|---|
| Certificate | C7A884AFCD7F91408921890D9617BEF6BFAC3DAE992BC25BB9C48C0395A60842 | TLS certificate on 38.244.50[.]120, issuer organization ARTEX | Source report only |
| Docker image | artex-local:perf2-20261003 | Custom image tag | Source report only |
| Domain | api[.]anthropic[.]com | Legitimate LLM service, not attacker infrastructure. Use only as an outbound behavior lead. | Shared or legitimate service |
| Domain | api[.]deepseek[.]com | Legitimate LLM service, not attacker infrastructure. Use only as an outbound behavior lead. | Shared or legitimate service |
| Domain | api[.]x[.]ai | Legitimate LLM service, not attacker infrastructure. Use only as an outbound behavior lead. | Shared or legitimate service |
| Domain | direct[.]xcai[.]pro | Endpoint in the model-routing configuration | Platform: suspicious |
| Domain | fofoapi[.]com | Third-party service, not attacker infrastructure. | Shared or legitimate service |
| Domain | open[.]bigmodel[.]cn | Legitimate LLM service, not attacker infrastructure. Use only as an outbound behavior lead. | Shared or legitimate service |
| Domain | u6ud2s[.]dnslog[.]cn | DNS callback hostname in a staged exploitation artifact | Platform: malicious |
| Domain | xcai[.]pro | likely LLM API proxy/reseller | Platform: suspicious |
| Fingerprint | 127.0.0.1:5435 | local PostgreSQL mapping | Source report only |
| Fingerprint | ARTEX - 自主渗透测试控制台(자율 침투테스트 콘솔) | HTML title on an attack-infrastructure web server (Section 4.2) | Source report only |
| Fingerprint | ARTEX — 自主渗透测试控制台 | ARTEX default title | Source report only |
| Fingerprint | LLM 驱动的自主渗透测试系统控制台(LLM 기반 자율 침투테스트 시스템 콘솔) | Meta description | Source report only |
| Fingerprint | grok2api | Grok bridge service | Source report only |
| IP | 101[.]53[.]80[.]20 | Proxy IP address | Platform: no verdict |
| IP | 103[.]248[.]148[.]84 | Proxy IP address | Platform: no verdict |
| IP | 124[.]155[.]252[.]63 | Proxy IP address | Platform: no verdict |
| IP | 154[.]201[.]79[.]246 | Proxy IP address | Platform: no verdict |
| IP | 161[.]8[.]161[.]140 | Upstream SOCKS proxy | Platform: no verdict |
| IP | 203[.]160[.]133[.]172 | Proxy IP address | Platform: malicious |
| IP | 205[.]214[.]59[.]31 | Proxy IP address | Platform: no verdict |
| IP | 209[.]209[.]85[.]38 | Proxy IP address | Platform: no verdict |
| IP | 23[.]158[.]220[.]98 | Proxy IP address | Platform: no verdict |
| IP | 23[.]248[.]226[.]250 | Work server | Platform: malicious |
| IP | 23[.]248[.]226[.]251 | Work server | Platform: malicious |
| IP | 23[.]248[.]226[.]252 | Work server | Platform: malicious |
| IP | 23[.]248[.]226[.]253 | Work server | Platform: malicious |
| IP | 23[.]248[.]226[.]254 | Work server | Platform: malicious |
| IP | 23[.]248[.]249[.]90 | Proxy IP address | Platform: malicious |
| IP | 38[.]244[.]50[.]120 | Threat actor-controlled IP address | Platform: malicious |
| File path | sync-memory.sh | custom synchronization script | Source report only |
| File path | tools/project_context_hook.py | context hook script | Source report only |
| File path | tools/sync-all.sh | custom synchronization script | Source report only |
| File path | tools/sync-cc-users.sh | custom synchronization script | Source report only |
| File path | xcai-shim.py | protocol shim | Source report only |
| URL | hxxp://38[.]244[.]50[.]120:18899/[.]claude/CLAUDE[.]md | Claude Code markdown file | Source report only |
| URL | hxxps://xcai[.]pro/v1 | ARTEX model gateway | Source report only |
Related events
Sources (32)
- Vendor Recorded Future Insikt Group TAG-217 Targets South Korean Financial Services with ARTEX
- 2026-10-07 Vendor CrowdStrike Counter Adversary Operations Unknown Threat Actor Uses AI-Driven ARTEX to Target South Korean Finance
- 2026-10-07 Vendor CrowdStrike Unknown Threat Actor Uses AI-Driven ARTEX to Target South Korean Finance
- 2026-10-09 Vendor Logpresso [로그프레소 CTI] OSINT를 활용한 ARTEX 캠페인 추가 분석 - 공격 집단 실체 검증
- 2026-10-12 Vendor Genians Security Center 금융분야 침해사고 이슈 분석 (Financial Sector Breach Incident Analysis)
- 2026-10-14 Vendor Zscaler ThreatLabz Threat Actor Abuses Agentic AI Pentesting Framework to Target South Korean Financial Institutions
- 2026-10-21 Vendor Zscaler ThreatLabz Inside a Global AI-Assisted Hacking Campaign
- 2026-10-29 Vendor AhnLab AI-Discovered Vulnerabilities: PHP, ARTEX and October's Cyberattacks
- 2026-11-04 Vendor Hunt.io ARTEX Exposed: Infrastructure Behind the South Korea Bank Hack
- 2026-11-05 Vendor Censys ARC ARTEX: Agentic Asbestos
- 2026 Media MIT AI Risk Repository / AI Incident Database Attackers reportedly used ARTEX, an AI security-testing tool (AI Incident 1782)
- 2026-10-01 Media 한국경제 'AI 에이전트'에 당했다…신한은행 보안 허점 노출
- 2026-10-02 Media 연합뉴스 국민·하나·부산은행도 당했다…금융권 해킹 피해 확산(종합2보)
- 2026-10-02 Media 한겨레 하나은행, 'AI 에이전트 활용' 추정 해킹 공격으로 89명 개인정보 유출
- 2026-10-02 Media 아이뉴스24 [은행 보안 구멍] 해킹 공격에 줄줄이 뚫려⋯최소 6.5만여명 개인정보 유출 [종합]
- 2026-10-03 Media 헤럴드경제 [단독] 은행 연쇄 해킹 도구는 중국 '아르텍스 AI'…해커가 사용했다
- 2026-10-03 Media SBS Biz 현대캐피탈도 해킹 피해…대출모집인 정보 일부 유출
- 2026-10-04 Media 연합뉴스 "금융사 해킹사고, 동일 공격 IP 여러 곳 발견…AI활용 추정"(종합)
- 2026-10-04 Media SBS Biz 웰컴저축은행도 정보유출…법인고객 정보 2천200건 추정
- 2026-10-07 Media 뉴스1 해킹도구 된 中 보안용 AI 'ARTEX'…"韓서 금융권 해킹 첫 사례"
- 2026-10-07 Media CEO스코어데일리 신한·국민·하나 등 금융권 연쇄해킹…'과징금 리스크' 실적 변수는
- 2026-10-08 Media The Hacker News ARTEX AI Pentesting Tool Used in Data Theft Attacks on South Korean Financial Firms
- 2026-10-08 Media The Korea Herald Chinese-speaking hacker possibly linked to AI-driven attacks on S. Korean banks: report
- 2026-10-09 Media Cloud Security Alliance Labs ARTEX: Open-Source Agentic Pentesting Tool Linked to South Korean Bank Intrusions
- 2026-10-10 Media BleepingComputer ARTEX AI, Claude agents used in cyberattacks on South Korean banks
- 2026-10-12 Media 주간경향 은행은 왜 AI해킹의 먹잇감이 됐나?
- 2026-10-13 Media SBS Biz 금감원, 예가람·웰컴저축은행 해킹 사고 현장 점검 착수
- 2026-10-20 Media 비즈워치 '해킹 42시간 몰랐던' 하나은행…'이상행위 자동차단체계' 구축
- 2026-10-23 Media 바이라인네트워크 신한은 15시간, 국민은 67시간 만에 알았다…금융권 국감서 해킹 늑장 인지 추궁
- 2026-10-26 Media 연합뉴스 해킹 발표 24일 만에…KB국민은행 정보유출 119→2천210명(종합)
- 2026-10-06 Secondary 중앙일보 [단독]현대캐피탈, 신한보다 먼저 털리고도 5일간 '깜깜이'…보안원 지적에 '첫 인지'
- 2026-10-08 Secondary 아이뉴스24 [단독] 개보위, 신한은행·예가람저축은행 해킹 사건 '중요사건' 지정