SEOUL, October 11 (AJP) - Weeks before hackers used artificial intelligence tools to steal customer data from seven South Korean financial firms, the sector's security agency had walked those firms' security staff through a similar AI attack on a domestic lender, according to documents the agency submitted to a lawmaker.
The breaches began in late September. They exposed data on about 40,000 Yegaram Savings Bank customers and about 25,000 Shinhan Bank customers. KB Kookmin Bank, Hana Bank, BNK Busan Bank, Welcome Savings Bank and Hyundai Capital also confirmed leaks. Woori Bank and NH NongHyup Bank said they blocked intrusion attempts.
The Financial Security Institute (FSI) submitted the material to Han Chang-min, who leads the Social Democratic Party and sits on the National Assembly's National Policy Committee. Han's office disclosed it Sunday. The documents show the agency described the earlier attack at a Sept. 17 seminar attended by about 160 security officers. Staff from four of the firms later breached attended, along with officials from Shinhan's holding company.
An AI agent is software that carries out a chain of tasks on its own after a single instruction. In the September case, the attacker told an agent built on the Chinese model DeepSeek to scan the financial sector for servers running Oracle WebLogic, software behind many corporate websites and apps.
The next day, the agent exploited a flaw that let outsiders run commands on one company's server and planted a web shell there. A web shell is a small malicious program that gives an intruder a back door. The company, identified only as Company A, stopped the intruder as they moved toward other internal servers, just before more malicious code could be downloaded.
The FSI urged companies to keep constant watch over their internet-facing systems. "Once- or twice-a-year assessments will not catch it," the presentation said. "Constant measurement and verification are needed."
The wider campaign came within two weeks. U.S. cybersecurity firm CrowdStrike said in an Oct. 7 report that the attacker used ARTEX, an open-source penetration-testing tool developed in China. Tools like it are meant to let companies probe their own defenses, and ARTEX hands much of that work to AI agents.
The copy of ARTEX that CrowdStrike examined ran on DeepSeek V4.1-Flash, and the attacker used Anthropic's Claude Code and other AI models in separate sessions. Files left on the attacker's servers, one of them in Hong Kong, showed the attacker asking Claude where stolen South Korean data is sold.
CrowdStrike assessed with moderate confidence that the attacker is a Chinese speaker motivated by money. It has not linked the activity to any known hacking group.
The targets were also different. Company A was breached through an unpatched product from an outside vendor. In the later campaign, the attacker went after services the banks ran themselves, including one bank's loan inquiry page for brokers and another's mobile work app for employees, CrowdStrike said, citing industry reports.
ARTEX's developer, known on GitHub as Autumn-27, said Oct. 8 that the tool "will no longer be updated" and would go closed source because it had been misused. The change does not affect versions already released.
Han blamed the companies themselves. "This leak of personal information at financial companies was caused by their failure to properly manage vulnerabilities that had already been stressed again and again," he said. He added that lawmakers would examine at the audit whether the companies responded properly before and after the breaches, and would hold them accountable.
The chief executives of South Korea's five largest commercial banks are scheduled to testify about the breaches at the National Assembly's audit of the Financial Supervisory Service on Oct. 19.
AJP Takeaways
- The Financial Security Institute walked about 160 financial-sector security officers, including staff from four of the seven firms later breached, through an AI-agent attack on a domestic lender at a Sept. 17 seminar, according to documents it submitted to Social Democratic Party leader Han Chang-min.
- CrowdStrike said in an Oct. 7 report that the attacker behind the breaches used ARTEX, a Chinese-developed open-source penetration-testing tool running on DeepSeek, along with Anthropic's Claude Code, and assessed with moderate confidence that the attacker is a Chinese speaker motivated by money. These findings come from CrowdStrike's report, not the lawmaker's documents.
- Han Chang-min blamed the companies for failing to manage vulnerabilities that had been flagged again and again, ahead of an Oct. 19 National Assembly audit where the heads of South Korea's five largest commercial banks are scheduled to testify.
Copyright ⓒ Aju Press All rights reserved.