Be Alert to AI Exacerbating Financial Cybersecurity Risks
Multiple banks and financial institutions in South Korea have successively experienced personal information leaks. Regulators assess that attackers used AI tools to automatically change IPs and scan for vulnerabilities in bulk, involving ARTEX and large language models assisting in executing some tasks. During the relevant period, a total of 16 financial institutions reported security incidents, approximately 100,000 customer information records were leaked, 555 institutions completed 12 security checks, and 130 require rectification. The article argues that financial cyberattacks are evolving from "humans using AI as assistance" to "AI agents autonomously taking on some attack tasks," and governance needs to be upgraded from individual institutional protection to national-level coordinated defense.
Body · AI translation
In early October, several banks and financial institutions in South Korea experienced successive personal information leakage incidents, and signs were found that attackers were using artificial intelligence (AI) to carry out attacks. South Korean President Lee Jae-myung ordered an investigation. South Korea's Financial Services Commission convened an emergency meeting with financial industry associations, regulatory agencies, and the financial institutions involved, requiring the entire industry to implement the highest level of alert and actively respond to this round of large-scale cybersecurity threats.
The above situation shows that, with the rapid improvement of AI agent capabilities, financial cyberattacks are showing a trend of evolving from "humans using AI to assist attacks" to "AI agents independently undertaking part of the attack tasks." The risk is not only enhanced attack capability, but also lower attack costs, faster attack speeds, and expanded attack scale, which confronts the traditional governance model, which mainly relies on individual institutions defending independently, with structural challenges.
Main characteristics of the leakage incidents
This data leakage incident involving South Korean financial institutions shows several main characteristics.
First, the institutions involved cover a wide range, and many important financial institutions have encountered cybersecurity threats. Multiple financial institutions, including Shinhan Bank, KB Kookmin Bank, Hana Bank, and Busan Bank, successively reported personal information leakage or cyberattacks. During the relevant period, a total of 16 financial institutions reported security incidents, involving banking, savings banks, credit, and digital finance, with about 100,000 customer information records leaked. Fortunately, no leakage of information directly related to illegal payment risks, such as account passwords or bank card numbers, has been found.
Second, the attack behavior shows the characteristics of batch-based, automated, cross-institutional scanning. South Korean regulatory authorities assessed that the attackers were not targeting a single institution for pinpoint strikes, but were scanning system vulnerabilities at multiple financial institutions over a wide area. The South Korean government required financial institutions to carry out comprehensive security inspections, tighten access permissions, and reduce unnecessary external system connections. An official notice on October 14 showed that 555 institutions in the financial industry had completed 12 security checks, of which 130 reported issues requiring rectification, mainly involving security patches and monitoring systems.
Third, technical evidence of AI-assisted attacks has further strengthened. Media speculated that the attackers used AI tools to continuously change Internet Protocol (IP) addresses and automatically attack multiple financial companies. The related activities are suspected of using ARTEX and large language models to assist in executing some tasks.
Taken together with this South Korean incident and the first AI-related personal information leakage case disclosed recently in Singapore, the dual risks facing the global financial industry—externally AI-enabled attacks and internally non-compliant AI use—are rising. The evolution of attack methods, superimposed on insufficient threat information sharing among institutions, makes the issue of joint defense in the financial industry more prominent.
First, cyberattacks are gradually evolving from "humans using AI to assist attacks" to "AI agents independently undertaking part of the attack tasks." Traditional cyberattacks place high demands on attackers' abilities in coding, vulnerability discovery, and system penetration; AI can assist in interpreting code, analyzing system configurations, helping discover vulnerabilities, and automatically generating attack scripts. With the rapid development of AI agents capable of autonomously invoking browsers and code tools, the technical threshold for conducting large-scale cyberattacks has dropped significantly. The South Korean incident suggests that the possibility of AI undertaking some automated tasks in specific attack activities has gained more concrete support from security research.
Second, the attack pattern is shifting from single-point targeted attacks to large-scale wide-area scanning attacks, and the "barrel effect" of the financial ecosystem has become more prominent. This South Korean incident reflects that cyberattacks show the characteristic of batch scanning and attacking multiple institutions. As the capabilities of AI tools and agents improve, attackers may further conduct automated vulnerability detection against common technical components, network interfaces, and third-party systems of different financial institutions. Once common vulnerabilities appear in these systems, attacks will be launched against a large number of institutions in a short period.
Third, after data leakage, AI amplifies secondary harm and intensifies the risk of precision fraud. In the past, after criminals stole massive amounts of personal data, they needed to expend a large amount of manpower on screening and analysis; AI can quickly complete data cleaning, correlation comparison, user profiling, and target screening. Once AI fuses stolen financial data with social, e-commerce, communications, and publicly available corporate data, it can precisely lock onto high-value targets and automatically generate highly customized phishing emails, text messages, and fraud scripts, further threatening the property safety of the public.
Fourth, non-compliant use of AI tools within institutions creates new data leakage risks. The case previously disclosed in Singapore shows that, although its scenario differs from South Korean financial institutions suffering external attacks, employees bypassing internal approval to use public AI tools may cause sensitive corporate data to leave the existing security management system. For financial institutions, externally AI-assisted attacks and internal "shadow AI" use constitute two types of risks that need to be prevented simultaneously.
The traditional security governance model, with individual financial institutions as the main responsible entities, faces new challenges. On the basis of the existing financial cybersecurity and data security systems, it is necessary to further strengthen coordination mechanisms across regulatory departments, financial institutions, and industries, with a focus on improving real-time sharing of AI attack threat intelligence, cross-institutional risk warning, joint response, intelligent defense, and the ability of the financial system to maintain business continuity under extreme circumstances, promoting the upgrade of financial security governance from "individual institutional protection" to "national-level coordinated defense."
First, establish a national joint response mechanism for financial AI security. It is recommended to study and establish a national joint response mechanism for financial AI security, incorporating banks, securities, insurance, fund, and payment institutions as well as key fintech service providers into a unified threat intelligence collaboration system, so as to achieve "one institution detects, the whole industry is warned." For key threat intelligence such as high-risk attack characteristics, malicious IPs, and vulnerability exploitation methods, promote the sharing timeliness from the traditional weekly and daily levels to the hourly and minute levels, and simultaneously establish intelligence desensitization and confidentiality management mechanisms as well as data security compliance requirements.
Second, promote key financial institutions to build an intelligent security operations system that "uses AI to defend against AI." Drawing on the "use AI to defend against AI" concept of South Korea's Financial Services Commission, promote large banks, securities companies, insurance institutions, and key payment infrastructure to build AI security operations centers. Rely on AI means to continuously monitor network traffic, abnormal access, account behavior, and data flows, use intelligent agents to identify potential attack risks and output disposal plans, and promote the transformation of financial security from rule-driven passive defense to intelligent active defense.
Third, establish an AI red team and normalized offensive and defensive testing mechanism for the financial industry. Establish a normalized AI red team mechanism for systemically important financial institutions. On the premise of strictly fulfilling compliance authorization and clearly defining testing boundaries, use controlled offensive intelligent agents to simulate real attacks, conduct automated security testing on internet-exposed assets, network interfaces, employee accounts, third-party access systems, internal AI applications, and database permissions, and identify potential security vulnerabilities through normalized simulated attacks.
Fourth, strengthen security supervision of financial institutions' external systems and third-party supply chains. Drawing on South Korea's approach in this incident of compressing external access and strengthening access control, improve China's unified security assessment system for financial institutions' AI tools, cloud platforms, software services, outsourcing service providers, data service providers, and interfaces, and strictly implement zero trust and the principle of least privilege. Implement continuous security assessments for third-party entities that can access customer information and core business data, and prevent attackers from bypassing financial institutions' core systems and carrying out cyber intrusions through peripheral supply chains.
Fifth, improve the financial industry's AI tool access and "shadow AI" governance system. Urge financial institutions to establish a complete AI inventory, and clearly define the business purposes, users, data permissions, and full-process data flows of large models, AI agents, and code generation tools used within the institutions; for AI applications involving sensitive customer information, core business data, and important data, prioritize secure and controllable technical architectures such as private deployment and dedicated clouds. Measures in Singapore's regulatory practice such as AI tool management, permission control, input and output monitoring, and full-lifecycle testing of AI development can serve as references for Chinese financial institutions to implement.
Sixth, incorporate AI-related cyber risks into financial security stress tests and emergency drills. Organize normalized practical emergency drills to simulate extreme scenarios such as "multiple banks simultaneously suffering AI automated attacks," "large-scale leakage of financial customer data," and "AI-generated false financial information inducing market fluctuations," and test the coordinated response capabilities of regulatory authorities, financial institutions, telecom operators, and cybersecurity agencies.
(The author is dean of the Zhejiang Zhijiang Digital Economy Innovation Research Institute and chief expert of the ASEAN Country Research Institute of Shaoxing Institute of Technology)