A 26-Year-Old Amateur Exposed South Korea’s Banking Data Using AI

A 26-year-old man living in southern China allegedly breached the financial systems of at least nine major South Korean banks, exposing the data of tens of thousands of residents. The intrusion, which cybersecurity firm CrowdStrike identified as the work of an amateur, relied heavily on artificial intelligence chatbots to execute the attack and market the stolen information.

Investigators pieced together the perpetrator’s identity, age, and location by analyzing digital traces and the failure to secure his own systems. The attacker used AI tools to draft a fake resume touting cybersecurity skills and to help sell the data online. CrowdStrike found that the operation utilized both Anthropic’s Claude model and free models from Chinese firms, including DeepSeek, because Anthropic’s tools generally refuse to perform illegal actions.

The incident has intensified concerns that AI is accelerating the pace and reach of cyberattacks. Adam Meyers, senior vice president of counter adversary operations at CrowdStrike, noted that the technology allows non-experts to leverage complex frameworks against targets. He stated, “Anyone can be a super-hacker now.”

In April, Anthropic restricted access to its new Mythos AI model to trusted partners and the government, deeming it too effective at hacking for general release. The move prompted experts to use the model to uncover thousands of dormant vulnerabilities. The Trump administration subsequently placed restrictions on how Anthropic and OpenAI could release their newest models, limiting access to a short list of government-approved firms.

Business leaders have criticized the administration’s approach, arguing that restricting access to top-tier models leaves companies vulnerable. Malte Ubl, chief technology officer of start-up Vercel, said the delay cost the industry three to five months during which threats could have been addressed. He noted that when AI agents from OpenAI hacked the company Hugging Face, the firm had to rely on free open-weight models to investigate.

Last month, the U.S. National Security Agency advised websites to shut down if they could not immediately patch a major security hole in PHP, a programming language used by an estimated 70 percent of websites. The vulnerability was discovered by a research firm using AI tools. Jason Clinton, Anthropic’s deputy chief information security officer, warned that the situation would deteriorate before it improved, stating that the industry needs to hold on until AI helps write code without security vulnerabilities.

The accessibility of powerful AI tools is increasing as free open-weight models improve. Anthropic said in September that the best free models now perform close to the level of Mythos when it was first released in April. Will Wilson, chief executive of Antithesis, a start-up that uses AI for code security, called the idea that frontier labs could keep this capability restricted “laughable.”

Recent breaches have underscored the growing role of AI in digital theft. Billions of dollars in digital assets were stolen from cryptocurrency exchange Binance last week. Nick Carlsen, a senior investigator with TRM Labs, said that while there is no specific evidence of AI use in that hack, the technology likely explains the recent increase in sophistication by the North Korean group behind the heist.

Cybersecurity researcher Cathedral reported on Monday that they accessed internal messages and meeting transcripts while testing Anthropic’s security. Anthropic executive Logan Graham responded to the findings on X, writing, “I’m concerned this is what the future will look like. Adversaries will use whatever models/tools they can. They’ll target everything with agents 24/7.”

Snehal Antani, chief executive of Horizon3, said critical infrastructure organizations like energy and water companies need immediate and comprehensive testing. He urged for resources to fund remediation and a public-private partnership to accelerate incident response. The FBI and Environmental Protection Agency said in August that malicious cyber actors have been remotely tampering with equipment at water plants, causing loss of pressure and flooding.