A Chinese AI tool was used in attacks on South Korean banks

For the damage, CrowdStrike relies on industry reporting, and it leaves the number of affected organisations open. At one bank the attacker reportedly got into a service financial brokers use to check how a loan is progressing. At another, into a mobile work system for staff. The New York Times writes that personal credit information of thousands of customers leaked at several banks. President Lee Jae Myung said signs had emerged that AI models were used in some of the attacks, and that this was causing considerable public concern. The South Korean police told the Times they would not comment while the investigation was underway.

The Register, citing The Korea Times, names at least five institutions: Shinhan Bank, KB Kookmin Bank, Hana Bank, Yegaram Savings Bank and BNK Busan Bank. For three of them it gives a figure: about 25,000 customers at Shinhan, 119 at KB Kookmin and 89 at Hana. I have not checked those figures with the banks or with The Korea Times. The Register also writes that police are looking into whether this was one person or a group, and that parliament wants to hear the heads of five major banks on 19 October.

What CrowdStrike read itself was sitting in open folders on the attacker's servers: Claude Code sessions, ARTEX configuration, and Claude memory files. In those sessions the user asked Claude where people sell stolen Korean data, and for help finding Korean Telegram groups that sell it.

In one session the user asked Claude for a security researcher's résumé, with the ARTEX results written into it. The prompt included a name, YY, a phone number, an age of 26 and a birth date in 2007, plus a Chinese university and a city in Guangdong. The age and the birth date contradict each other. CrowdStrike says the details likely belong to the person using the session, and that it cannot tie them firmly to the attacker. When The New York Times called the number, a man answered in Chinese and said he was not YY.

Personal credit information of bank customers. The Times says thousands of people. The Register, via The Korea Times, puts Shinhan alone at about 25,000. CrowdStrike has not confirmed how many organisations were hit, or how much was taken.

This is a different shape from the break-in at Hugging Face and the Australian government portal. There the agents belonged to a lab, and they got out of a test environment. Here a person pointed programs that were already available, Claude Code among them, at banks.

The next day Axios uses this campaign as the example of what one person can do, and it writes more firmly than CrowdStrike that the hacker is from China. That piece is here: the day after a major incident. CrowdStrike's own conclusion is that adversaries will keep trying AI tools, so they can move faster and do more.

I have read the CrowdStrike report, the New York Times piece and The Register. I have not opened The Korea Times, which is where The Register gets the bank names and the customer counts. I have also not opened the Korean article CrowdStrike cites in a footnote.

I did not call the phone number, and I am not printing it here. Nor the university, nor the city. CrowdStrike publishes them, and says in the same breath that the link is not established.