South Korea's financial sector is under sustained cyberattack. Between late September and October 29, 2026, at least 19 financial institutions — spanning banks, brokerages, asset managers, and cryptocurrency exchanges — have confirmed data breaches or unauthorized access attempts. The cumulative ...
"It is deeply troubling that firms are still failing to meet basic security requirements despite repeated calls for comprehensive inspections and stronger controls." — Shin Jin-chang, Secretary General, South Korea Financial Services Commission (FSC)
South Korea's financial sector is under sustained cyberattack. Between late September and October 29, 2026, at least 19 financial institutions — spanning banks, brokerages, asset managers, and cryptocurrency exchanges — have confirmed data breaches or unauthorized access attempts. The cumulative exposure exceeds 100,000 customer records.
The latest wave, disclosed October 28, hit crypto exchange Bithumb (6,444 user records), brokerage Daishin Securities (38,490 customers), and Mirae Asset Global Investments ($461 billion AUM). These follow an earlier cluster that compromised seven banks — including Shinhan Bank, KB Kookmin Bank, and Hana Bank — via ARTEX, an open-source AI-driven autonomous penetration testing framework developed in China and released on GitHub in July 2026.
CrowdStrike and Recorded Future have independently attributed the earlier campaign to TAG-217, a likely Chinese-speaking, financially motivated threat actor who combined ARTEX with multiple commercial LLMs — including DeepSeek, Grok, and Zhipu AI — to automate reconnaissance and exploit web-facing systems at scale. The Financial Services Commission (FSC) has responded by proposing twice-yearly security audits, stricter executive accountability, and a unified cybersecurity framework that treats crypto firms and banks under the same regulatory perimeter.
The attacks have arrived in three distinct clusters:
Wave 1 — Late September to Early October: TAG-217 used ARTEX and LLM-assisted reconnaissance to target seven South Korean banks and financial institutions, including Shinhan Bank, KB Kookmin Bank, Hana Bank, Hyundai Capital, Welcome Savings Bank, BNK Busan Bank, and Yegaram Savings Bank. The actor exploited authentication and access-control gaps in loan-brokerage query services, employee mobile business-support systems, and customer-facing portals. According to Recorded Future's Insikt Group, TAG-217 had been assembling tools and testing infrastructure as early as August 17, 2026, five weeks before the first public disclosures.
Wave 2 — October 2–14: Crypto exchanges Gopax and Prabang were breached. Prabang reported an October 2 cyberattack resulting in data leaks. By October 14, the Korea Times reported that 16 financial organizations had confirmed incidents, with an estimated 100,000 records exposed across all affected entities.
Wave 3 — October 17–29: Daishin Securities detected unauthorized access to a server supporting remote account openings on October 17. Bithumb confirmed unauthorized access to its internal mobile groupware system (Daou Office) on October 27. Mirae Asset Global Investments disclosed abnormal outside access to business-use web systems on October 28. These three disclosures, arriving in the same 48-hour window, brought the total count of confirmed incidents to at least 19 firms.
ARTEX — an open-source, LLM-driven autonomous penetration testing system — sits at the center of the first attack wave. Developed by Autumn-27 and released on GitHub on July 26, 2026 under the GNU AGPL v3.0 license, ARTEX won Baidu's "agent+" attack-defense challenge before being deployed against live financial targets.
Architecture: ARTEX uses a multi-agent architecture comprising a Planner module, multiple Worker agents, an Engine scheduling loop, and a MITM (man-in-the-middle) proxy. It is written in Go with a Next.js frontend and is deployable via Docker or a single binary.
LLM Stack: According to CrowdStrike's October 7 analysis, the TAG-217 instance used DeepSeek v4.1-flash as its primary LLM backend, supplemented by Zhipu AI's GLM-5.3 and SpaceX AI's Grok 4.6 for task-specific functions. Recorded Future additionally identified Claude Code among the AI tools present in the threat actor's working environment.
Attack Methodology: The multi-agent system automated asset discovery, vulnerability identification, and exploitation verification. It used residential proxies to obscure geographic origin and probed web-facing systems across loan-brokerage services, mobile business platforms, and customer portals. CrowdStrike assessed with moderate confidence that the operator was Chinese-speaking and financially motivated, based on Chinese-language prompts found in session logs and the tool's Chinese origin.
Closure: On October 8, following public attribution, Autumn-27 announced that ARTEX would go closed-source. The original GitHub repository was taken down. However, source code backups and Korean-localized forks remain available on the public internet, according to Recorded Future.
The broader significance: ARTEX represents the operationalization of AI-assisted offensive tooling. A single operator, working with freely available open-source tools and commercial LLMs, breached seven regulated financial institutions in under two weeks. The FSC's self-assessment subsequently found that 130 of 555 financial organizations (23.4%) had cybersecurity deficiencies — most commonly delayed security updates, unsupported legacy systems, and the absence of multi-factor authentication on critical services.
The three most recent breaches, disclosed October 28–29, illustrate distinct attack vectors:
South Korea's largest crypto exchange by trading volume confirmed that unauthorized parties infiltrated its internal Daou Office mobile groupware system on October 27. Of the 6,444 leaked records, 5,600 were encrypted and deemed unreadable by Bithumb's assessment. The remaining 844 records were fully exposed, along with 715 employee or partner personnel files. Compromised data categories include names, phone numbers, email addresses, resident registration numbers, driver's license numbers, bank account details, and copies of identification documents.
Bithumb stated that no abnormal withdrawals or customer asset losses occurred, as the Daou Office system is isolated from trading and custody infrastructure. The exchange reported the incident to Korea's Personal Information Protection Commission and the Korea Internet & Security Agency (KISA).
Daishin detected suspicious access on October 17 to a server supporting remote securities-account openings for discretionary investment services. The breach exposed eight categories of personal and credit information for 38,490 customers: names, email addresses, residential addresses, communication IDs, bank names, deposit and withdrawal account numbers, securities account numbers, and mobile phone numbers. Daishin blocked the offending IP addresses and reported to regulators.
Mirae Asset, with $461 billion in assets under management as of September 2026, reported unauthorized access to certain business-use web systems. The firm stated that retail investor information was not involved, as the affected systems do not serve those customers. No financial losses were reported, and core systems continued operating normally.
Korean authorities noted that North Korean IP addresses appeared in connection with some of the recent activity, but cautioned that IP provenance alone does not establish attribution. The investigation remains ongoing.
Aggregating across all three waves:
| Metric | Figure | |---|---| | Total confirmed breached firms | 19 | | Estimated records exposed | 100,000+ | | Banks breached (Wave 1) | 7 | | Crypto exchanges breached | 3 (Bithumb, Gopax, Prabang) | | Brokerages/asset managers breached | 2 (Daishin, Mirae Asset) | | Other financial firms with reported incidents | 7 | | Firms with self-reported deficiencies (FSC audit) | 130 of 555 (23.4%) | | Primary attack tool | ARTEX (AI pentesting framework) | | LLMs used by attacker | DeepSeek, Grok, Zhipu AI, Claude Code | | Attribution | TAG-217 (likely Chinese-speaking, financially motivated) | | Attack preparation start | August 17, 2026 | | First public disclosure | Early October 2026 | | Most recent disclosure | October 28–29, 2026 |
No customer asset theft has been confirmed at any of the affected firms to date. The primary damage is data exfiltration — personal information, credit data, and identification documents — which carries long-term identity fraud risk.
According to a Cisco survey of 8,000 security executives across 30 countries cited by Seoul Economic Daily, 97% of South Korean companies report lacking adequate security personnel. EY Hanyoung Financial AI Center's Lee Gun-young noted that "most financial institutions are complaining of a shortage of AI security personnel, as the network separation environment has left insufficient systems for financial security staff to grow or be trained."
The Financial Services Commission moved on multiple fronts:
1. Mandatory Self-Assessment: By October 8, the FSC required all 555 regulated financial organizations to evaluate their systems against a 12-point cybersecurity checklist. The 23.4% deficiency rate (130 firms) revealed systemic gaps, particularly among electronic financial service providers and P2P lending platforms.
2. Twice-Yearly Security Audits: The FSC proposed increasing cybersecurity assessment frequency from annual to biannual for large financial institutions. This includes crypto exchanges, which would be subject to the same schedule as banks — a notable regulatory convergence.
3. Executive Accountability: Proposed rules would strengthen personal accountability for senior executives at firms that fail cybersecurity assessments, though specific penalties have not been detailed.
4. AI-Based Threat Monitoring: The FSC is considering mandating AI-powered security monitoring for smaller financial enterprises that lack the resources for 24/7 human-staffed security operations centers.
5. Emergency Inspection Authority: Regulators would gain authority to conduct supplementary security inspections outside routine evaluation cycles during periods of elevated threat.
6. Unified Crypto-Bank Perimeter: The most structurally significant proposal: South Korea would treat banks and crypto firms as part of the same security perimeter, applying identical cybersecurity standards to both categories. This goes beyond the approach of most jurisdictions, which maintain separate regulatory tracks for traditional finance and digital asset firms.
7. ISMS Certification Review: FSC Secretary General Shin Jin-chang stated: "We take seriously the criticism that the ISMS is ineffective, and we will re-examine whether the certification framework and standards are appropriate." FSC Chairman Lee Eog-weon is scheduled to meet with chiefs of major financial industry associations and CEOs of the country's five commercial banks on November 5 to review sector-wide defenses against AI-enabled cyberattacks.
The South Korean attack wave carries three implications for the broader market:
AI lowers the cost of financial-sector attacks. A single operator, using freely available tools and commercial LLM APIs, breached seven banks in under two weeks. The marginal cost of probing additional targets approaches zero once the AI agent is configured. This represents a structural shift in the threat landscape — not because AI creates new vulnerability classes, but because it compresses the reconnaissance-to-exploitation timeline from weeks to hours.
Crypto-TradFi regulatory convergence accelerates under pressure. South Korea's decision to apply identical cybersecurity standards to banks and crypto exchanges was driven by the fact that both were hit in the same campaign. When attackers do not distinguish between financial subsectors, regulators are forced to follow. This pattern may replicate in other jurisdictions experiencing cross-sector cyber incidents.
Data breaches at crypto firms carry asymmetric risk. Bithumb's breach did not touch custody or trading systems, but the exfiltrated data — resident registration numbers, bank account details, ID copies — provides the inputs for targeted phishing, SIM swapping, and social engineering attacks against crypto holders. The secondary attack surface created by data exfiltration is larger in crypto than in traditional finance, because crypto custody often relies on individual key management where a single successful social engineering attack can result in irreversible asset loss.
South Korea's five-week cyberattack wave is the first documented case of an AI-driven pentesting tool being used at scale against a major economy's financial sector — hitting both traditional banks and crypto exchanges in the same campaign. The 68,000 records exposed in the first wave, plus 45,000+ in the latest disclosures, demonstrate that AI-assisted offensive tooling has moved from proof-of-concept to operational deployment.
The FSC's response — treating crypto firms and banks as a single security perimeter — is a regulatory precedent with potential global implications. As financial infrastructure converges (stablecoin rails at banks, bank custody at crypto firms), the attack surface has already converged. South Korea's experience suggests that cybersecurity regulation will follow.
ARTEX went closed-source on October 8, but forks and backups persist online. The tool's architecture is well-documented. The capability it represents cannot be unbuilt. The question for every financial jurisdiction is not whether AI-assisted attacks will arrive, but whether defenses will be in place when they do.