Published: October 8, 2026 · Last updated: October 8, 2026
TL;DR: CrowdStrike says the person behind a wave of AI-assisted attacks on South Korean banks is likely a financially motivated individual based in China’s Guangdong province, probably around 26 years old. The firm says he combined Anthropic’s Claude Code, an AI coding agent, with an open-source Chinese tool called ARTEX to automate parts of the intrusion. Investigators reportedly got closer to his identity partly because he used an AI assistant to help write his own resume.
CrowdStrike published a report this week identifying the person likely behind a string of AI-assisted bank breaches in South Korea as a financially motivated hacker probably based in Guangdong province, China, and probably around 26 years old. The detail getting the most attention is the toolkit: the firm says the attacker ran Anthropic’s own Claude Code, an AI coding agent built for software engineers, alongside an open-source Chinese tool called ARTEX to automate pieces of the intrusion.
This builds on a story we covered this week, South Korea’s president saying AI tools had helped an attacker breach Shinhan Bank, exposing data on roughly 25,000 customers after the attacker bypassed identity checks and cycled through query values. Other reporting on the president’s warning put the full campaign at seven banks total, with data on about 68,000 people exposed before it was shut down. CrowdStrike’s new report is the first detailed attribution work tied to that campaign.
Korean bank hacks have historically pointed toward North Korea. CrowdStrike tracks a state-linked group it calls Ricochet Chollima that has gone after South Korean financial and government targets for years, usually for hard currency or intelligence. This one reads differently, according to the firm: a financially motivated, Chinese-speaking individual working largely alone, not a state operation. Investigators reportedly narrowed in on the suspect partly through an unforced error. He had apparently used an AI assistant to help write his own resume, and identifying details leaked through in the process.
The Claude Code detail is worth sitting with. Anthropic built the tool for engineers writing and debugging software, not for breaking into banks, but an agentic coding tool doesn’t check who is doing the prompting. CrowdStrike says it handled scripting and automation that would have taken a skilled operator real time to write by hand. This also isn’t the first time Anthropic’s own tooling has turned up on the wrong side of an intrusion. The company disclosed last November that a separate, state-sponsored Chinese group had used Claude Code to automate an espionage campaign of its own. Pair that history with ARTEX, an open-source Chinese agentic framework built for offensive automation, and the setup here needed far less custom development than a comparable operation would have five years ago.
That is the pattern security researchers keep flagging this year: agentic AI tools are shrinking the gap between someone with real hacking skill and someone who can just describe what they want done. It tracks with other moves we have seen recently, Google pausing part of its open source bug bounty program this month because AI-generated vulnerability reports were burying its human reviewers, and enterprise security startups raising hundreds of millions specifically to control what AI agents are allowed to touch inside corporate systems.
None of this is settled law enforcement fact yet. CrowdStrike’s report is an attribution assessment built on technical fingerprints and operational security slip-ups, not a confirmed identity backed by an arrest, and South Korean authorities have not said whether they have located or charged anyone. Treat the age, the province and the resume detail as CrowdStrike’s current best read of the evidence, not a closed case.
Related: This builds directly on South Korea’s president saying AI helped hackers breach Shinhan Bank, and it lands the same week Google paused part of its bug bounty program because AI-generated reports were drowning out real vulnerability submissions.
Bottom Line: The useful part of this story isn’t the suspect’s age or hometown, it’s the toolkit. A financially motivated hacker working mostly alone reportedly ran a legitimate AI coding agent next to an open-source attack framework and got most of the way through a multi-bank breach before a resume-writing habit gave him away. That’s the actual threat model security teams need to plan for now. Not a nation-state budget, just someone with API access and a plan.
Follow Teck Hustlers: Facebook · X · Instagram · LinkedIn · TikTok · Pinterest
