‘Hacking for just $5?’ Chinese AI Agent Targets South Korean and Japanese Firms

【Financial News, Tokyo—Correspondent Seo Hye-jin】An AI agent that identifies security vulnerabilities in systems appears to have been misused in cyberattacks targeting companies in South Korea and Japan. The attacks involved using AI for everything from finding vulnerabilities to breaking into systems. One analysis also found that the average cost of using AI in attacks targeting online shopping sites in Japan, the United States and elsewhere was just $5 per attack.
According to The Yomiuri Shimbun on the 20th, an investigation by Japanese security firm GMO Cybersecurity by Ierae raised the possibility that a Chinese-language AI agent called ARTEX was used in attacks on about 80 Japanese companies. The targets spanned industries including insurance, finance and logistics, and eight of the companies disclosed that they had suffered information leaks.
While conducting a joint investigation with an external organization at the request of affected companies, GMO found six overseas public servers used in attacks on Japanese companies between the 18th of last month and the 18th of this month. Traces of ARTEX use were found on all six servers. With no signs that other AI agents had been used, GMO concluded that ARTEX was likely involved in the attacks.
ARTEX is a program made public in July this year by a developer who uses Chinese, on an online service for sharing technology information. It can broadly identify vulnerabilities in systems. An AI agent is a system that can carry out multiple steps of a task on its own after receiving a simple instruction. In the security field, AI agents are used to check for vulnerabilities and gather information, among other things.
The same tool was also used in attacks targeting South Korean financial institutions. According to U.S. security firm CrowdStrike, ARTEX was used in attacks on multiple South Korean financial institutions in September and October this year. The developer said it had confirmed that the tool had been misused in attacks and halted its public release on the 8th of this month. However, the developer maintained that it was “designed to strengthen the security of companies and organizations.”
JPCERT/CC, Japan’s cyber incident response organization, also said it had identified traces of AI agent use in attacks targeting Japanese companies since August this year.
Separately, U.S. security firm Noforge announced on the 19th (local time) the results of an investigation showing that Chinese-made software called Cairn had been misused in attacks targeting e-commerce sites in Japan, the United States and Australia. Cairn is open-source software that activates multiple AI agents in response to simple instructions.
An analysis of Cairn’s management interface by Noforge found that attackers instructed AI to autonomously identify vulnerabilities in websites and break into them. Investigators found records showing that an AI agent had exploited vulnerabilities to obtain unauthorized control privileges. Records also showed that illegal commands had been executed on a Japanese retailer’s website.
The attacks began sometime last month or later, and several Japanese websites were among the targets. However, it has not been confirmed whether customer information was actually leaked from the Japanese retailer in question. Noforge raised the possibility that AI from China’s DeepSeek was used. The attack instructions were written in Chinese and English.
The main attacks took an average of about 23 hours. AI usage costs, including those for failed tasks, were estimated at an average of about $5 per attack. While these costs should be distinguished from the overall cost of the attacks, the findings point to vulnerability searches and intrusions being automated for a low AI usage fee.
Daiki Fukumori of GMO, who led the investigation, said, “The affected companies were attacked through minor system vulnerabilities,” adding, “Companies should conduct a comprehensive review to make sure they have not overlooked any vulnerabilities.”
[email protected] Seo Hye-jin Reporter