CrowdStrike, the U.S. cybersecurity company, said on Wednesday, Oct. 7, that an attacker who targeted South Korean banks and other financial firms, some of them hit in a recent run of data breaches, made heavy use of AI tools. In files the attacker left open on the internet, CrowdStrike found session records from Claude Code, Anthropic's AI tool that can write code and run commands on a computer, along with settings for a free AI security-testing tool called ARTEX. It says the campaign ran from late September to early October and that data was taken.
ARTEX is an AI agent, software that can carry out tasks on its own instead of only answering questions. It is meant for penetration testing, where security teams attack their own networks to find weak spots, and it is open source, free for anyone to use, study, change and share. Its own project page says it should not be used to test real online systems, according to a news report. ARTEX has no AI of its own; it plugs into AI models made by other companies. CrowdStrike says the attacker's copy ran mainly on a model from the Chinese AI company DeepSeek, and that the attacker also ran Claude Code on models from Zhipu AI, another Chinese company, and Elon Musk's xAI.
The breaches have exposed bank customers' personal details. According to a news report, at least nine South Korean banks have disclosed attacks or been reported as targets since late September, and South Korean police opened an investigation this week. The same report says Shinhan Bank put the number of customers whose personal information was compromised at about 25,000, and KB Kookmin Bank at 119. It is not clear which of these breaches were this attacker's work, and CrowdStrike says the number of organizations affected is still unconfirmed.
CrowdStrike says the attacker also asked Claude where stolen Korean data is usually sold, and for help finding Telegram groups that trade it. The files hold personal details that CrowdStrike says likely belong to the attacker, though it says it cannot confirm that. It has not tied the attacks to a known hacking group, and it judges with moderate confidence that the attacker speaks Chinese and was after money, based on the attacker's use of a Chinese-made tool and the Chinese-language instructions it found.
CrowdStrike's warning is about speed: it says AI tools let one attacker carry out several break-ins in a short time, and it expects attackers to keep trying them out. Anthropic did not respond to a request for comment, according to the news report, and has not said whether it spotted or stopped any of this activity. CrowdStrike has not said how much faster the attacks went with AI than they would have without it.