CrowdStrike ties five Korean bank hacks to one man using AI agents
One attacker used Claude Code and a Chinese AI hacking tool on five Korean banks, then left his logs and résumé exposed.
CrowdStrike says the hacks that hit five South Korean banks between late September and early October were most likely the work of one person running AI agents. The banks were Shinhan, KB Kookmin, Hana, BNK Busan and Yegaram Savings. At Shinhan alone about 25,000 customers were affected, and Infosecurity Magazine reports around 40,000 at Yegaram.
The main tool was ARTEX, a free, open-source AI pentesting agent built in China. Pentesting means probing a system for weak spots, which is normally done with permission. He paired it with Claude Code, given hacking instructions written in Chinese, and switched between several models behind the scenes, with DeepSeek's cheap Flash model doing most of the work. The agents found holes in things like a loan progress lookup page and a mobile system bank staff use for work, and got in through them.
He was caught by his own mess. One of his servers had a folder left open to the internet, and in it were his AI session histories, config files and the memory files his agents kept about each target. CrowdStrike also found a résumé, apparently drafted with AI help, listing a Chinese university and a location in Guangdong, plus a Telegram username that turns up elsewhere in the campaign. The details don't fully add up, since the résumé says he is 26 but gives a birth date in 2007, and CrowdStrike says it can't prove the résumé belongs to the attacker. It rates the overall attribution as moderate confidence.
The fallout is now political. South Korean lawmakers have voted to summon the heads of the five banks to a parliamentary audit on October 19 to explain how this happened.
What stands out to me is how ordinary the setup was. No state hacking team, no custom malware, just a free agent off GitHub, a couple of AI subscriptions and cheap models, pointed at public bank websites until something gave. A few years ago five major banks in a few weeks would have meant a group of skilled people. Here it looks like one guy who couldn't keep his own server locked.
The other side of that is worth noticing. The same agents that made him fast also wrote down everything he did, target by target, in files he forgot to hide. AI makes attackers quicker, but it also makes them leave a much more detailed paper trail.
My actual takeaway is about the banks. The pages that got hit were the boring ones, a loan status checker and an internal staff app, the kind nobody thinks of as the front door. Agents are very good at patiently checking every one of those. If one person with free tools can work through five banks this quickly, the bar for basic security just went up for everyone.