Successive infringements by Shinhan, Kookmin, Hana, and Busan banks... We and Nonghyup were also attacked
Hacking formula changed by AI agent... Defense with AI and Zero Trust
(Seoul = Yonhap News) Reporter Kwon Ha-young = There is an interpretation that the series of hacking incidents that hit the banking sector are not a one-off incident, but are the result of the realization of risks that were predicted in the process of artificial intelligence (AI) moving from a 'tool' to an 'agent' that makes decisions and executes on its own.
As attackers form organizations and AI designs and executes the infiltration process that used to be performed one by one by hand, even banking institutions, which were considered to have the strictest security, are being breached one after another in a short period of time.
◇ Following Shinhan Bank, Kookmin, Hana, and Woori Bank suffered hacking incidents in succession.
According to the financial sector on the 4th, after it was confirmed that the personal information of about 25,000 customers was leaked from Shinhan Bank through a loan originator service on the 30th of last month, external infringements were revealed one after another at KB Kookmin Bank, Hana Bank, and BNK Busan Bank around the same time.
Many of these incidents have in common that they were subjected to repetitive and automated external attacks at a level that was difficult for humans to respond to manually, and it is presumed that personal information of customers, employees, or outsourced personnel was leaked in the process.
At KB Kookmin Bank, the information of 119 customers was leaked through the employee mobile work system, and at Hana Bank, the information of 89 customers was leaked through the sales support system. At BNK Busan Bank, during an external web server attack attempt, the information of 11 outsourced development employees was exposed through a page where session verification was insufficient.
Woori Bank and NH Nonghyup Bank also suffered similar attacks, but it was reported that it did not lead to an actual information leak.
Given that several financial companies were simultaneously targeted by attacks in a short period of time, concerns are growing that this may be a structural vulnerability shared by the financial sector as a whole, rather than a loophole in a specific bank.
◇ “Why now?”… Foreshadowed risks brought about by AI autonomy
The background to this incident lies in the stages of development that AI technology has gone through.
Unlike the generative AI of the past, which was limited to answering people's questions or finding necessary information, recent AI is evolving into the form of an 'agent' that sets its own goals and performs several steps in a chain.
This change is evaluated as having changed the ‘performer’ of the attack rather than the attack technique itself.
In the past, in order to find loopholes by inputting large amounts of information, people had to manually perform repetitive tasks or write separate automation scripts, but now, by simply specifying a goal for an AI agent, it can collect information, search for vulnerabilities, execute attacks, and even modify methods according to the results.
Google Cloud and its security subsidiary Mandiant warned in their 'Cybersecurity Outlook 2026' report that attackers will make full use of AI.
The representative methods pointed out in the report are 'prompt injection', which secretly inserts malicious instructions into AI to trick it into doing the wrong thing, and 'AI-based social engineering (social engineering fraud)', which deceives the opponent by elaborately manipulating voice, text, and video with AI.
The report defined this trend as an 'AI agent paradigm shift' and cited the impact this would have on overall security as a key threat.
East Security, a domestic security company, also predicted that as AI is used as a key tool in all stages of an attack, an environment will be created where ordinary attackers can attempt attacks at a level that only professional organizations could carry out in the past.
For this reason, there is an analysis that this banking incident may be a “security failure of a specific bank,” but it is also “a rite of passage that was scheduled to coincide with the development stage of AI technology.”
◇ Defense with AI… Point out that “there are limits to borderline security”
Security experts agree that it is difficult to keep up with the speed and scope of AI-led attacks with the existing method of manually responding to AI-led attacks.
Until now, the security system has focused on preventing intrusions at the border dividing the outside and inside by blocking only the entrance to the building.
However, the common diagnosis in the industry is that in a situation where attackers use AI to search for vulnerabilities 24 hours a day, at a much faster rate than humans, this perimeter-based defense alone is bound to leave a gap.
For this reason, there are growing voices in the industry that defense systems should also use AI to monitor attack surfaces at all times and detect abnormal signs in real time.
There are also suggestions that, rather than completely blocking the intrusion itself, the Zero Trust principle should be applied to tightly control the flow of the internal network to prevent access to core assets even if an intrusion occurs.
The security industry also emphasizes that since this incident was not a defect in a specific system, but occurred at a point where different services and data are connected, an integrated security check that goes beyond individual vulnerability checks and encompasses the entire organization's work flow is necessary.
Financial authorities are said to be considering ways to reorganize the security inspection system across the banking sector using this incident as an opportunity.
