South Korea is investigating a string of attacks on financial institutions after authorities found signs that artificial intelligence may have played a role in some of the intrusions. The development is significant not simply because AI was allegedly involved, but because the incidents appear to have targeted the less visible systems that connect banks with employees, brokers and other third parties.
President Lee Jae Myung said on October 6 that investigators had found indications of AI use in some recent hacking incidents and called for the circumstances to be established quickly. South Korean authorities have launched a wider investigation, while financial regulators are asking institutions to check externally exposed systems and strengthen their defenses. Reuters reported that the Financial Supervisory Service had shared information on 28 unique IP addresses associated with the recent attacks. Authorities have not publicly established exactly which AI systems were used or how much of the attack activity was automated.
The reported breaches have affected several financial companies. Shinhan Bank said information relating to around 25,000 customers was exposed, while other reports have described incidents involving KB Kookmin Bank, Hana Bank and additional financial firms. Woori Bank and NH Nonghyup Bank were also targeted, although unauthorized access was blocked in those cases. Yonhap reported that investigators had identified 28 IP addresses across 12 countries and territories, while The Korea Times reported that the leaked Shinhan information included names, phone numbers and annual income.
Importantly, this does not mean that an AI system independently broke into a bank. That distinction matters. At this stage, authorities are investigating evidence suggesting that AI tools may have helped attackers discover weaknesses or carry out parts of the operation. The exact tooling, attack chain and level of automation have not been publicly established by investigators.
The more interesting part of the story is where the attackers looked
Early reporting suggests that the attackers were not necessarily trying to punch through the heavily protected core systems used to move money. Instead, attention has focused on peripheral and third-party systems such as employee support services, loan-broker portals and information lookup tools.
That is an important cybersecurity lesson. A modern bank is not a single application protected by a single perimeter. It is a network of internal applications, vendors, cloud services, contractors, APIs, identity systems and portals. A system that looks secondary from the perspective of the customer can still contain enough information to become valuable to an attacker.
One reported example involved a portal used by loan brokers to track customer applications. The Financial Times reported that an attacker bypassed identity checks on such a system, exposing information associated with roughly 25,000 Shinhan Bank customers. The report also said investigators believe attackers may have focused on third-party systems rather than the banks’ core payment networks.
This is where the AI angle becomes more concerning. Finding a forgotten internet-facing application, identifying its technology stack, testing authentication weaknesses and moving from one exposed service to another are tasks that traditionally required considerable time and expertise. Reporting from The Korea Times says investigators found traces associated with ARTEX AI, an open-source autonomous penetration-testing system, on infrastructure believed to have been used in the attacks. Its presence is evidence worth investigating, but it does not by itself prove exactly how the tool was used. AI-assisted tools can potentially compress reconnaissance and testing, allowing attackers to test more targets in less time.
AI could change the economics of an attack
The biggest change may not be that AI makes an individual hacker dramatically more capable. It may be that AI makes large-scale experimentation cheaper.
An attacker who previously had to manually inspect dozens of systems can increasingly use automation to classify targets, generate requests, inspect responses, identify likely weaknesses and decide which systems deserve human attention. Even when an AI agent cannot complete an intrusion on its own, it can reduce the amount of repetitive work surrounding the attack.
That creates an uncomfortable imbalance. Defenders already have thousands of alerts, logs, endpoints and software dependencies to monitor. Attackers only need to find one weak external service. If AI reduces the cost of looking for that weak point, the defender’s workload can increase much faster than the size of the security team.
There is another reason to be careful with the headlines. Calling something an “AI hack” can make an incident sound more autonomous than it actually was. AI can be used for reconnaissance, code generation, vulnerability research, social engineering or decision support without being the component that ultimately gains access. Until investigators publish the technical details, it is too early to describe these incidents as autonomous AI attacks.
What banks should take from this
For financial institutions, the immediate lesson is not to build a bigger wall around the core banking platform and assume the problem is solved. The perimeter now includes every internet-facing service that can reach sensitive information or trusted internal systems.
That means security teams need a continuously updated inventory of external assets, including systems operated by vendors and contractors. Authentication should be treated as a critical security boundary rather than a box that can be checked once during a compliance review. Logs from smaller business applications should receive the same attention as logs from customer-facing systems when those applications handle personal or financial data.
It also means defenders need to think about AI differently. If attackers can use automation to search for weaknesses, defenders should use automation to discover exposed services, correlate suspicious activity and prioritize remediation. The Korea Times reported today that the Financial Supervisory Service instructed financial firms to identify externally accessible infrastructure, assess vulnerabilities and review authentication and authorization controls. South Korea’s financial authorities have also been pushing the sector toward AI-assisted defenses.
But automated defense should not become an excuse for weaker fundamentals. Patch management, strong identity controls, least-privilege access, segmentation, vendor security reviews and accurate asset inventories remain the foundations. AI can make those processes faster. It cannot compensate for an organization that does not know what it has exposed to the internet.
For ordinary customers, there is no evidence from the reports cited here that money was stolen directly through these incidents. However, leaked personal information can create a second wave of risk. Names, phone numbers, income information and other identity details can be useful for phishing, impersonation and targeted fraud. South Korean financial authorities have warned that secondary damage such as voice phishing and smishing cannot be ruled out.
Our take
The most important story here is not “AI hacked banks.” It is that the boundary between a bank’s core infrastructure and the many smaller systems surrounding it is becoming increasingly difficult to defend.
For years, cybersecurity conversations have focused heavily on protecting the crown jewels: payment systems, databases and customer applications. The South Korean incidents are a reminder that attackers may prefer a side entrance. A loan portal, contractor account or internal lookup service may not look as important as online banking, but if it provides a path to customer data, it becomes part of the bank’s security perimeter.
AI makes that problem more urgent because it can potentially turn reconnaissance from a slow, selective process into a high-volume activity. The right response is not panic about autonomous hackers. It is better visibility, faster patching, stronger identity controls and continuous testing of everything an organization exposes to the outside world.
And there is a final caveat worth keeping in mind: the investigation is still underway. The evidence may ultimately show that AI played a narrow supporting role rather than driving the attacks. That distinction will matter, because understanding exactly what the attackers automated is what will tell defenders which parts of today’s security model need to change.
Sources and further reading: Reuters, Financial Times, Yonhap News Agency, and The Korea Times. This article adds editorial analysis and context based on the publicly reported facts available on October 6, 2026.
