AI agents hack South Korean banks exposing 25,000 customer records
The AI Arms Race: How Hackers Are Turning Intelligent Models Against the World’s Biggest Banks
A new and alarming chapter in cybersecurity is unfolding as sophisticated cyberattacks utilizing AI agents target some of South Korea’s largest financial institutions. Investigations are underway regarding breaches at major banks, including Hana Bank, KB Kookmin Bank, and Shinhan Bank, highlighting a growing vulnerability in secure financial systems.
The fallout from these incidents has already affected customer data. Last week alone, reports indicated that at least 25,000 Shinhan Bank customers had their personal credit information compromised. Similarly, customers and employees of Kookmin Bank, as well as 89 customers of Hana Bank, were among those whose information was stolen. This data breach underscores a critical fact: the sophisticated tools of modern hacking are now being augmented by artificial intelligence.
The severity of the situation has drawn the attention of national leaders. President Lee Jae Myung has publicly addressed the crisis, noting that there are signs suggesting the attackers employed AI models. He stressed that this development has caused considerable public concern and anxiety, and he has ordered the cabinet to confirm the situation and minimize the resulting damage. “It’s now become possible to use AI to hack with ease even without specialized skills,” the President stated during a recent cabinet meeting, signaling a shift in how we view digital defense.
This threat is not theoretical; it is rooted in the accelerating power of AI itself. As AI models become more advanced with every release, malicious actors are exploiting their potential to find weaknesses in secure systems and launch attacks with unprecedented efficiency. The recent banking incidents are a stark demonstration of how AI agents can be weaponized for cyber warfare.
The history of AI-powered cyberattacks reveals a dangerous trajectory. Reports from cybersecurity firms have detailed how AI has been used to orchestrate complex operations, including the first instance of an AI-orchestrated cyberattack allegedly originating from a Chinese state-sponsored group in late 2025. Furthermore, AI has facilitated end-to-end autonomous attacks, demonstrating the potential for these systems to execute sophisticated strategies in real-time.
The vulnerability extends even to the developers of these powerful models. Security researchers have faced breaches themselves, demonstrating that the very systems designed to protect the public are not entirely immune. In one notable incident, researchers gained access to internal codebases and employee accounts by breaching platforms like OpenAI using sophisticated tools. This highlights a significant challenge: ensuring the security of the AI tools themselves.
The current exploitation strategy focuses on the inherent weaknesses of AI systems. Attackers are leveraging phenomena like ‘hallucinations’—where AI models dream up false information—to trick agents into executing malicious code. This exploit turns the AI’s tendency to generate plausible but incorrect outputs into a potent tool for running malware.
Compounding the issue, many open-weight and open-source models lack the robust security guardrails found in proprietary systems, making them easier targets. While AI models haven’t replaced the need for highly skilled human attackers, they act as a powerful force multiplier. They drastically reduce the time required for planning, reconnaissance, and execution, allowing bad actors to act with speed and efficiency previously unimaginable.
As the digital landscape continues to evolve, the challenge for global security is no longer just about building better defenses, but about understanding and securing the intelligent tools that now drive the attack. The future of cybersecurity will depend on bridging the gap between AI development and impenetrable defense.
Buy on Amazon
Credit
