Category: Security | Reading time: 4 min
What Happened
Since late September, seven South Korean financial firms have confirmed intrusions: Shinhan Bank, KB Kookmin Bank, Hana Bank, BNK Busan Bank, Yegaram Savings Bank, Welcome Savings Bank and Hyundai Capital. Woori Bank and NH NongHyup Bank detected signs of hacking attempts but have not confirmed leaks.
The Wall Street Journal reported on 6 October that officials put the number of people whose personal data was stolen at about 68,000. The attacks showed traces of ARTEX AI, an open-source, LLM-based autonomous penetration-testing system distributed mainly through GitHub to Chinese-speaking users.
The Korea Financial Security Institute told The Herald Business it traced the attack IPs and server logs at Shinhan Bank, the first institution to report, and found evidence pointing to ARTEX. The institute was explicit that a person was in charge: "AI was used in the attacks, but the AI did not act independently without human involvement. A hacker used the AI as a tool."
Where the Attackers Went In
None of the intrusions hit internet or mobile banking. Every reported breach came through internal systems used by employees, contractors or loan agents:
- Shinhan Bank: 25,729 people's data exposed through a loan-agent inquiry service, including names, phone numbers, annual income and loan limits. - KB Kookmin Bank: 119 records from an employee mobile work-support system. - Hana Bank: 89 records from an employee sales-support system. - BNK Busan Bank: information on 11 contract workers.
According to The Korea Times, ARTEX can connect to models from DeepSeek, OpenAI and Anthropic and run multiple agents to map targets and intrusion routes. Authorities said the attackers rotated through IPs in many countries, including the United States, Japan, Hong Kong, Singapore, Vietnam, Thailand and the United Kingdom. Because the tool is public and the infrastructure is spread out, regulators said they cannot yet attribute the campaign to a country or group.
Hwang Suk-jin of Dongguk University described it as entering "through a window" rather than the front door: support networks have weaker controls than core banking systems. He warned the breaches may have been an initial test for larger follow-up attacks.
The Regulatory Response
The response has been sector-wide and fast:
- Emergency self-inspections. The Financial Supervisory Service shared attacking IPs and advisories with about 500 financial firms. Banks and card companies had to finish checks of externally exposed IT assets, access controls and patch status by 6 October; securities firms, insurers, savings banks and electronic financial operators by 8 October. The FSS has now circulated 28 distinct IP addresses linked to the attempts. - Deregulation paused. The Financial Services Commission shelved the 7 October selection for the second round of its emergency network-separation relaxation, which had been run since June to allow AI security testing and was set to expand to 75 eligible firms. The FSC said the agenda needed additional security review. - Employee-facing systems. The FSC plans to recommend that financial firms run a comprehensive audit of internal employee-facing systems among their external access points. - Criminal probe. The Korean National Police Agency's cyber terror unit opened a full investigation on 6 October.
Authorities said voluntary remediation of basic IT controls will run through November, with stern action if large breaches follow from inadequate inspections. FSC chair Lee Eog-weon told an emergency meeting that the sector must "quickly establish security systems that defend AI attacks with AI."
Why It Matters
This is one of the first confirmed cases of an agentic offensive tool showing up in intrusions across a national banking sector, and the regulator's first move was to slow down a program that loosened controls for AI. Two lessons carry beyond Korea.
First, the soft targets were the side systems: loan-agent portals, sales-support tools, staff mobile apps. These are exactly the internal surfaces where enterprises are now wiring in their own action-taking agents.
Second, the questions regulators are asking after the fact are about evidence: which systems were exposed, who had access, what was patched and what each actor did. When both attackers and defenders run agents, institutions need a record that separates their own agents' authorised actions from everything else, and that record has to be ready before the examiner asks for it.
Source: The Wall Street Journal | The Herald Business | The Herald Business (FSC pause) | The Korea Herald | The Korea Times | Chosun Biz