One attacker, a whole banking sector: AI-assisted hacking in South Korea

At a glance: In a report published on 7 October 2026, CrowdStrike linked attacks on South Korean financial organisations from late September to early October to one threat actor using ARTEX, an open-source AI penetration-testing tool, alongside several large language models. Shinhan Bank says about 25,000 customers' data was exposed. CrowdStrike assesses with moderate confidence that the actor is likely a Chinese speaker and financially motivated. Boards should reassess the scale one attacker can now reach.

What happened

From late September to early October 2026, attackers reached systems at several South Korean financial institutions and took data. At least nine banks have been targeted since late September.

Shinhan Bank disclosed that personal data of about 25,000 customers had been exposed through a platform that loan brokers use to store customer information during loan applications. The bank said an unauthorised external party reached some services through "an abnormal method that bypassed authentication". The exposed data included names, phone numbers, annual income and loan eligibility limits. Chief executive Jung Sang-hyuk apologised, saying: "I sincerely apologize for causing concern and inconvenience to our customers." The bank said it would take full responsibility for any losses, blocked external IP addresses, suspended affected services and set up a dedicated call centre.

The Financial Supervisory Service carried out an on-site inspection of Shinhan Bank, and the FSS and the Financial Services Commission held an emergency meeting on the breach. South Korean police opened a formal investigation on 6 October covering seven financial firms.

What CrowdStrike found

On 7 October, CrowdStrike published its analysis. It has not attributed the activity to any named adversary. It describes the actor as "likely a Chinese speaker and financially motivated", and states: "This assessment is made with moderate confidence." That assessment rests on the use of a Chinese-developed tool and on Chinese-language prompts it observed. CrowdStrike says the number of affected organisations "remains unconfirmed". WCRC does not go beyond that description.

The tooling is the story for boards:

• ARTEX, which CrowdStrike describes as "a recently released open-source agentic penetration testing (pentesting) tool developed in China".
• Large language models behind it. CrowdStrike names a DeepSeek model as the primary back end, supplemented by Zhipu AI's GLM and xAI's Grok in additional sessions run through Anthropic's Claude Code, a coding assistant. A Chinese-language prompt in Claude Code configuration files directed the models' penetration-testing work.

CrowdStrike says it found the actor's AI session histories, tool configurations and memory files in openly accessible directories on servers the actor ran. The chats also included questions about where Korean breach data is typically sold. At one bank, according to industry reporting CrowdStrike cites, the actor breached a loan progress inquiry service used by brokers; at another, an employee mobile work-support system.

CrowdStrike's own assessment is that adversaries will likely keep experimenting with AI tooling to increase their operational tempo.

The other side of the same shift

A few days earlier, Google paused new product vulnerability submissions to its Open Source Software Vulnerability Reward Program, effective 1 October 2026. Google said: "This pause is due to a significant rise in automated submissions, the vast majority of which are not valid." Reports submitted before 1 October are unaffected, researchers are directed to other Google reward programmes, and Google said it would update on the area in the first quarter of 2027.

In WCRC's analysis, the two events are linked. AI is lowering the cost of producing attacks and of producing security reports. On the attack side, the Korean case suggests one operator can run reconnaissance and exploitation across many targets with automated help. On the defence side, Google found automated submissions, most of them invalid, rising sharply in a channel built for human researchers. Both point to volume as the new problem.

What changes in board assumptions

Many board cyber-risk discussions rest on assumptions about attacker capacity: that a sector-wide campaign needs a well-resourced group, that attackers must choose targets carefully, and that a breach at a peer gives time to prepare. The Korean case, as described by CrowdStrike and the banks, challenges each of these. This is WCRC's interpretation, not a finding of any investigation.

Scale is cheaper. An agentic tool that automates testing lets a small operation probe many organisations in parallel. Boards should assume that a weakness common across a sector may be found and tried across the sector at once.

Peripheral systems matter. The systems named, a broker-facing loan inquiry service and an employee mobile support tool, are not core banking platforms. Partner portals, third-party channels and staff tools are attractive because they are numerous and often less hardened.

Warning time shrinks. If campaigns run at machine speed, the gap between a breach at one bank and an attempt at the next narrows. Sector information-sharing has to be faster to be useful.

Defence is also flooded. Google's pause is a reminder that AI-generated noise can swamp intake processes. Security operations, bug bounty programmes and fraud teams will need their own automation to triage.

For banks in Asia and the Gulf

Banks across Asia and the Gulf share features with the Korean sector: dense digital channels, broker and agent networks, and heavy use of mobile tools for staff. In WCRC's view, boards in those markets should ask management for a short review, focused on:

• An inventory of internet-facing services used by partners and brokers, with authentication strength for each.
• Detection tuned for high-volume, automated probing rather than only targeted intrusion.
• Pre-agreed customer notification and support plans, including a channel for customers to check exposure, as Shinhan set up.
• Faster sector-level sharing of indicators with peers and regulators.

What it means for leaders

• Update threat models: one actor with AI tools can now attempt sector-wide campaigns.
• Prioritise partner portals, broker platforms and staff mobile tools, not just core systems.
• Assume a breach at a peer means you are next within days, not months; rehearse accordingly.
• Invest in automated triage for security reports and alerts; AI noise is rising too.
• Ask your security team how AI is used in your own defence, and how fast it can respond.

Quick answers

Q: Who was behind the South Korean bank hacks? A: CrowdStrike has not attributed the activity to a named group. It assesses with moderate confidence that the actor is likely a Chinese speaker and financially motivated.

Q: What AI tools were used in the South Korean bank attacks? A: CrowdStrike says the actor used ARTEX, an open-source agentic penetration-testing tool developed in China, with models from DeepSeek, Zhipu AI and xAI, run in part through Anthropic's Claude Code.

Q: Why did Google pause its open-source bug bounty? A: Google paused new product vulnerability submissions to its Open Source Software Vulnerability Reward Program from 1 October 2026, citing a significant rise in automated submissions, most of which were not valid.