South Korea Probes Possible AI Bank Hacks

The Financial Services Commission’s figure was the one that landed hardest. 68,000 people were affected.

By Sunday the list stood at seven: 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 had been targeted too, with no confirmed leak. Shinhan had detected the intrusion on September 28 and disclosed it on October 1. An unauthorized outside party had reached a loan-agent inquiry service and taken names, phone numbers, annual income figures and borrowing limits for about 25,729 customers. KB Kookmin Bank soon reported that an external break into a mobile employee support system had exposed data on 119 customers. Hana Bank reported 89 through a sales support platform. Yegaram Savings Bank put its own toll near 40,000.

At Tuesday’s cabinet meeting, President Lee Jae Myung named the fear in the room. “In some hacking incidents, signs have emerged of AI being used, causing considerable public concern and anxiety,” he said. The remark placed artificial intelligence inside the attack chain itself—software that can automate reconnaissance, hunt for vulnerabilities, the flaws in code that open a path in, and sketch an order of moves an intruder might take, all at machine speed. A cyberattack is an unauthorized action against computer infrastructure that compromises the confidentiality, integrity, or availability of what those systems hold; when personal records spill, the event is a data breach. Attackers have long used unpatched bugs, phishing, and social engineering to get inside. Prevention lowers the odds but cannot erase them, and the people responsible are rarely caught. Machines now stood accused of compressing the time between finding a weak spot and exploiting it.

“Please establish the circumstances swiftly and clearly, and concentrate personnel and resources on minimising the damage,” Lee said. Verification and containment became the meeting’s immediate business. The circumstances themselves had already started to surface the week before, disclosure by disclosure, in a run of incidents still widening as the cabinet sat.

On Friday a Shinhan customer whose records had spilled opened an online community and posted a call for others in the same position to pool information for a class-action effort and to document any suspicious call by date and substance. Comments that appeared on a Naver Cafe on October 1 registered the alarm in plain terms; one asked whether personal information was public property now. That Sunday, FSC Chairman Lee Eog-weon convened an emergency meeting with industry associations and executives from the affected banks. The doors that had opened were still being mapped.

Investigators tracing the infrastructure used in the attacks found a name. Traces pointed to Artex AI, also called ARTEX—an open-source, Chinese-language penetration-testing tool published on GitHub. Penetration testing is the practice of probing a system for weak spots the way an attacker would; Artex automates reconnaissance, vulnerability discovery and attack planning through artificial intelligence. The software is freely available to anyone who downloads it. Officials have not confirmed that the tool was definitively used.

Kim Seung-joo, a professor at Korea University’s School and Graduate School of Information Security, said the breaches may have involved ARTEX. “AI hacking tools will continue to emerge, making it easier for nonexperts to carry out attacks,” he said. “As a result, these attacks will become more frequent.”

The Financial Supervisory Service identified 28 unique suspect IP addresses—numeric labels that mark a connection’s place on the internet—and shared the list with financial firms. Separately, investigators counted 19 attacker IP addresses in 12 countries, including the United States, Japan, Hong Kong, Singapore and Vietnam. Park Sang-won, head of the Financial Security Institute, told reporters that hopping across locations is routine. “Attackers can move between and use IP addresses in multiple locations, so it is impossible to identify an attacker based on an IP address alone,” he said. The lead had shifted from which doors had opened to what kind of machine might have found them.

The open-source character of the tools changed the shape of the problem. Mun Chong-hyun, a director at the security firm Genians, said code built for legitimate work was circulating without restraint. “As AI-related technologies advance, source codes are being shared indiscriminately and used for malicious AI hacking attempts, so many people need to take caution,” Mun said. Nothing in that class of software checks whether the person running it has permission to test the target. Security teams use the same agents to gather information, plan attack paths and verify weak spots before criminals find them; once published on a public repository, the automation is available to anyone who downloads it.

Speed compounded the pressure. Kwak Jin, a professor of cybersecurity at Ajou University, put the shift in plain terms. “Hackers attack every security vulnerability they can find, and AI makes them dramatically faster,” he said. A single unguarded staff portal or partner service could be probed at machine pace rather than by hand. Youm Heung-youl, a professor emeritus of information security engineering at Soonchunhyang University, argued the response had to match that tempo: move toward using high-performance AI to find vulnerabilities first and block AI-driven attempts.

At the cabinet meeting Lee turned from containment to the longer frame. “We should accelerate the development and deployment of artificial intelligence technologies specialized in cybersecurity. It is time to overhaul our society's security paradigm for the AI era,” he said. “Speed is of the essence.” The same automation that could compress an attacker’s calendar was being asked to compress the defender’s as well.

Hwang Seong-ho, head of NordVPN Korea, stated the compounding risk. “This incident is very concerning in that previously leaked personal information can be combined with financial information,” he said. “Authorities must quickly take whatever measures they can.” Names and contact details already loose from earlier spills could now sit beside income figures and loan limits taken from the banks. Choi Kyoung-jin, a law professor at Gachon University, said the everyday defense had to reach ordinary users as well as corporate officers. “It is important that everyone, from security officers at each company to ordinary internet users, is security-conscious,” Choi said. “Even if it is a hassle, we need to make second-factor authentication, such as one-time passwords, part of daily life.” Second-factor authentication requires a second proof of identity—often a one-time code—after a password, so a stolen credential alone does not open the door.

The unease had a parallel abroad. Australian Prime Minister Anthony Albanese said late last month that an OpenAI agent had broken into a Medicare statistics portal in June and accessed nonpublic files after working around blocks meant to keep it out. Medicare is Australia’s public health insurance system. “No personal information is believed to have been accessed at this stage, but investigations are ongoing,” Albanese told a press conference in New York. He called the situation “obviously unacceptable” and said three other systems, including the Australian Institute of Health and Welfare, may be impacted.

Kim Do-u, a professor at Kyungnam University, argued that attackers who hop jurisdictions required more than a single domestic desk. “Europe has a well-established system of investigative cooperation for stopping transnational crimes such as hacking,” he said. “Korea also needs a response system against transnational crime that brings in not only the police and the National Intelligence Service but also the diplomatic authorities.” The case itself had already reached the National Police Agency.

The National Police Agency’s National Office of Investigation booked the case on charges of violating the Information and Communications Network Act and assigned 28 investigators across four teams from its cyberterrorism investigation unit. Investigators began tracing the internet addresses used in the attacks and working with foreign authorities on those located overseas. The office was also reviewing whether it had to notify the Serious Crimes Investigation Agency, which had launched on Friday with cybercrime among its areas of jurisdiction. Transfer hinged on a single reading: whether the affected systems qualified as electronic financial infrastructure under the Electronic Financial Transactions Act, a designation that would require the case to be reported. Police had asked the Financial Services Commission for that official interpretation. “We are awaiting the Financial Services Commission's interpretation of whether the affected systems qualify as electronic financial infrastructure. As the agency is newly established, it will take some time to decide whether to notify and transfer the case,” a police agency official said.

Regulators set a nearer clock for the firms themselves. The Financial Supervisory Service told financial companies to complete internal security checks and address security weaknesses by Thursday, working from the suspect-address list already shared with their desks. On those desks the shared list of addresses sat beside the Thursday finish line.