AI agent linked to South Korean bank hacks goes closed-source

An investigation that gathered pace on 10 October 2026 showed how a freely downloadable AI 'penetration tester' called ARTEX was turned on tier-one South Korean banks. Between late September and early October a single, financially motivated operator used the autonomous agent to run the full attack lifecycle against Shinhan Bank, Yegaram Savings Bank and other financial firms, exposing the personal and financial data of more than 144,000 customers. CrowdStrike assessed with moderate confidence that the attacker was a Chinese speaker; the agent itself was built by Chinese cybersecurity engineer Li Puhua and chained together Anthropic's Claude Opus, OpenAI's ChatGPT and China's DeepSeek models. Under pressure, the developer converted ARTEX from open-source to closed-source on 8 October and took down its GitHub page. South Korean President Lee Jae Myung publicly confirmed that AI agents had played a central role in the breaches and ordered a multi-agency probe. The episode is being read as the first large-scale, publicly documented bank intrusion driven mainly by an autonomous AI agent, and prompted a round of warnings from Western regulators and vendors about agentic cybercrime.