Security people have been warning about AI hackers at conferences for years, usually with a slide deck and a stern face. Now it has happened. Hackers used an AI hacking agent to breach at least seven South Korean financial firms and stole data on about 68,000 people, according to a Wall Street Journal report and the security researchers who followed it.
It is one of the first AI-powered cyberattacks on the global financial system. And the lesson is not the one you might expect.
How the AI hacking agent hit South Korean banks
Seoul officials spotted the attacks the week before the news broke. On Oct. 6, South Korea’s National Police Agency opened a formal investigation through its cyber terror unit, Breitbart reported.
The victim list is long. The Korea Herald names Shinhan Bank, KB Kookmin Bank and Hana Bank. BankInfoSecurity adds Yegaram Savings Bank, Welcome Savings Bank and Hyundai Capital. Reports also say the attackers targeted NH NongHyup and Woori, though neither has confirmed a breach.
Here’s the thing: the stolen data included some customers’ annual income and personal-loan limits. That is not a password dump, but scammers love exactly this kind of detail. It is the stuff that makes a fake phone call sound real.
Attacks on this scale are no longer limited to banks, either. Hackers recently turned the fridges at hundreds of French supermarkets into ovens.
ARTEX: the AI hacking tool behind the breaches
Investigators call the weapon ARTEX. A Chinese security engineer built this open-source AI agent to help companies find holes in their own networks. Think of it as a locksmith’s kit that someone else walked off with.
ARTEX is not an AI model itself. It borrows brains from models like Claude, ChatGPT and DeepSeek. So this is not purely a Chinese-tool story.
Infosecurity Magazine reports that CrowdStrike found the attacker used ARTEX alongside Anthropic’s Claude. BleepingComputer adds that DeepSeek, Zhipu’s GLM and Grok were part of the mix.
American AI tools, in other words, helped hit a foreign bank. That should get the attention of every AI company in the US, and it lands as the Pentagon says it has completely stopped using Anthropic’s Claude.
One AI hacker agent, many intrusions
This is the part that should keep security chiefs up at night. CrowdStrike says the case shows how AI tooling can let a single financially motivated actor run multiple intrusions in a short time.
Lim Jong-in, a professor at Korea University’s graduate school of information security, put it plainly. “It would be difficult for humans to carry out attacks this fast and on such a large scale across the financial sector in just a matter of days,” he told The Korea Times. “It is highly likely that AI was used.”
So who is behind it? CrowdStrike assesses with moderate confidence that the attacker is a Chinese speaker and financially motivated. It has not named an adversary.
Reuters reports the suspect may be a 26-year-old in Guangdong province. Police have not confirmed that. Meanwhile, the same Korea Times piece quotes a professor who warns against blaming Beijing just because a Chinese-language tool turned up.
The hacker left the receipts
Now for the irony. CrowdStrike found open directories on the attacker’s own servers. They were packed with Claude Code session histories, ARTEX configuration files and Claude memory files.
The attacker even asked Claude to help find Korean Telegram groups where people sell stolen data.
An AI hacking agent run by an operator with sloppy housekeeping. Shocking absolutely no one who has ever left a laptop unlocked.
Not the crown jewels, but still a warning
The compromised systems were employee and loan sales-support tools, not core banking infrastructure. CrowdStrike pointed to a loan inquiry service that brokers use at one bank. At another, the target was an employee mobile work-support system. So nobody’s account balance went missing.
Still, the speed matters more than the haul. The campaign ran from late September to early October. Investigators traced it to more than two dozen IP addresses in about a dozen countries, including the US, Japan and Germany.
President Lee Jae Myung pressed the banks at a cabinet meeting to tighten their defenses. In BankInfoSecurity’s translation, he said the country must build security that detects attacks in advance and blocks them before they land.
What an AI-powered cyberattack means next
The police investigation is ongoing, and officials are seeking international help to trace those IP addresses. Regulators are also expected to push new security rules for critical systems. A final count of affected firms and records is still pending, and the numbers vary by source, from roughly 66,000 to 68,000 people.
The question banks everywhere are now asking: if one person with an open-source agent can do this in days, what happens when the next one aims at a bigger target?
