One Hacker, a Swarm of AI Agents, and a Rough Month for Korean Banks

One Hacker, a Swarm of AI Agents, and a Rough Month for Korean Banks

The classic heist movie needs a crew: the hacker, the inside guy, the getaway driver, the one who says "I'm getting too old for this." According to CrowdStrike, a recent campaign against South Korean financial institutions needed exactly one person and a stack of AI tools. Ocean's Eleven has been downsized to Ocean's One, and the other ten are now chatbots.

An Autonomous Pentest Tool Goes Off-Script

CrowdStrike says a financially motivated, likely China-based operator hit multiple South Korean financial organizations between late September and early October 2026. The attacker used ARTEX, a recently released Chinese open-source tool that lets AI agents run penetration tests largely on their own, automating reconnaissance, vulnerability discovery, attack planning and verification. Reporting indicates personal information on roughly 68,000 people was exfiltrated, and confirmed breaches included one bank's loan-progress inquiry service used by brokers and another institution's employee mobile work-support system.

The operator mixed and matched models, with DeepSeek as the main engine plus Zhipu's GLM and xAI's Grok, alongside Anthropic's Claude Code. The investigation cracked open because the attacker left open directories on their own servers containing Claude Code session histories, ARTEX configs and memory files, effectively publishing their playbook prompt by prompt. ARTEX's developer has since closed the source and stopped updates, saying the tool was built for learning and research.

AI Didn't Invent the Holes, It Just Found Them Faster

Here's the part worth tattooing on every IT roadmap: the AI didn't conjure new zero-days. The vulnerabilities were already there. What agentic tooling changed is the tempo, letting a single operator investigate targets and run several intrusions within days, a workload that used to require a team. CrowdStrike's assessment is that adversaries are now operationalizing agentic AI to move faster and hit more targets in a shorter window.

The angle most people miss: the systems that fell weren't the bank vaults on the evening news. They were a broker-facing inquiry portal and an internal mobile work tool, the side doors and secondary apps that rarely get the same scrutiny as the flagship site. Automated attackers love exactly that kind of forgotten surface, because an agent doesn't get bored crawling through the 40th subdomain. And yes, there's some poetic justice in an AI-assisted attacker being undone by a misconfigured open directory. Even the robots can't save you from basic hygiene.

If attackers can now sweep your whole footprint in an afternoon, "we'll patch the minor stuff later" is officially an expired strategy.

That forgotten partner portal or legacy login page is exactly what automated scanners will find first, so let our team help you map and harden the side doors before someone else's agent does.

Source: Security Affairs