South Korean investigators say attackers used Artex, a Chinese-made, open-source AI agent, to break into at least seven financial firms and walk away with personal data belonging to 68,000 customers. The tool was built to find network weaknesses, yet in this case it appears to have become the instrument used to find and exploit them.<\/p>\n
Officials first detected the intrusions last week. The stolen records included some customers’ annual income and personal-loan limits, the kind of detail that makes convincing phone or message scams far more plausible. On Tuesday, South Korea’s National Police Agency opened a formal investigation through its cyber terror unit. Investigators traced the activity to more than two dozen internet addresses spread across roughly a dozen countries, including the United States, Japan, and Germany. No culprits have been identified, and authorities have said they are seeking international cooperation to track them down.<\/p>\n
Artex is not an AI model. It is an agent, software that draws on existing large models, including Anthropic’s Claude Opus, OpenAI’s ChatGPT, and China’s DeepSeek, to run cybersecurity tasks. It was designed to help organizations find vulnerabilities in their own networks, not to enable attacks. South Korean officials have not said which underlying models were used in the bank breaches. The link to Artex itself is still being investigated.<\/p>\n
A chatbot answers questions. An agent pursues a goal, chaining together research, planning, and software tools with little human supervision. In the hands of a defender, that means a tireless auditor testing a network for weak doors. In the hands of an attacker, it means a lockpick that never sleeps. Security professionals have lived with this duality for years. The open-source penetration-testing framework Metasploit and the commercial red-team tool Cobalt Strike were both built for authorized testers, and both became staples of criminal intrusion kits. Artex may be walking the same well-worn path, only faster.<\/p>\n
Artex had drawn attention before this case. In September, it won a contest sponsored by several Chinese tech companies that named it the top agentic AI system for both offensive and defensive cybersecurity work. After reports of the initial breaches surfaced, the developer updated the user guidelines to prohibit unauthorized intrusions, data theft, and other malicious use. A rule posted after the break-in is a lock bolted on after the burglary, and no policy page can stop code that anyone can download.<\/p>\n
Investigators believe the tool was misused to break into the banks and pull customer data, yet public proof remains thin. The Genians Security Center, a Seoul-based cybersecurity analytics firm, identified Artex’s potential involvement shortly after the breach came to light. A server used in the attacks was found carrying an HTML page title with a Chinese-language string associated with Artex, while official channels have offered no details about the perpetrators. The link to Artex is still being verified, and the case joins a growing string of AI-linked cyber incidents.<\/p>\n
At a cabinet meeting Tuesday, President Lee Jae Myung pressed banks to shore up their defenses, calling for the necessary measures to be implemented immediately. The Financial Services Commission chairman said authorities could not rule out the use of artificial intelligence and called for an “AI attacks defended by AI” approach. Regulators believe the attackers broadly scanned multiple firms for vulnerabilities rather than targeting a single institution, which suggests the weakest systems simply lost the lottery. South Korean cybersecurity stocks jumped as much as 30 percent on Tuesday as fear moved through the market.<\/p>\n
The incident joins a growing list of cases tying AI systems to breaches. Anthropic alleged last year that state-sponsored Chinese hackers used its AI technology to automate break-ins at roughly 30 targets worldwide, including corporations and foreign governments. In September, Australian officials said an OpenAI agent had infiltrated a government website, accessing both public and nonpublic files on the country’s healthcare-statistics portal. In recent weeks, Google’s Gemini model autonomously accessed the internet and hacked other companies during a test of its own cybersecurity capabilities.<\/p>\n
A tool that can map a network and find a vulnerability can just as quickly be turned against the network that deployed it. South Korea’s investigation and its required international cooperation will be watched closely by financial firms weighing how much autonomous software to allow inside their networks.<\/p>\n
Artex is an open-source AI agent built by a Chinese cybersecurity engineer. It is not an AI model itself; it draws on existing models such as Anthropic’s Claude Opus, OpenAI’s ChatGPT, and China’s DeepSeek to perform cybersecurity tasks. It was designed to help organizations find network vulnerabilities.<\/p>\n
Investigators say at least 68,000 customers had personal data exposed. The stolen information included some customers’ annual income and personal-loan limits.<\/p>\n
Officials traced the intrusions to more than two dozen internet addresses spread across roughly a dozen countries, including the United States, Japan, and Germany. No culprits have been identified. South Korea’s National Police Agency has opened a formal investigation through its cyber terror unit and is seeking international cooperation.<\/p>\n