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

An open-source AI agent designed for penetration testing has been pulled from public access after cybersecurity researchers linked it to a string of cyberattacks on South Korean banks. The tool’s Chinese developer made ARTEX closed-source on October 8, 2026, the same day its GitHub page was taken down.

The attacks, which struck between late September and early October 2026, breached internal systems at between seven and nine South Korean financial firms. Approximately 68,000 individuals had personal data exposed, including names, phone numbers, annual incomes, and loan limits. Shinhan Bank bore the worst of it, with around 25,000 customer records compromised.

What ARTEX is and how it was allegedly used

ARTEX was built as an autonomous AI agent for authorized security testing. The developer behind the project, identified by the GitHub handle Autumn-27 and named as Chinese cybersecurity engineer Li Puhua, created it as an open-source resource for the security community.

CrowdStrike, one of the cybersecurity firms that investigated the incidents, tied the cyber operations to a financially motivated individual in China. The firm estimates the operator to be a 26-year-old based in the country. According to their analysis, the attacker used ARTEX in conjunction with Anthropic’s Claude Code, effectively pairing an autonomous penetration testing agent with a powerful AI coding assistant.

The combination proved effective against internal banking systems rather than public-facing applications. That distinction matters. Public-facing systems, like login pages and customer portals, tend to have the heaviest security. Internal systems often rely on the assumption that anyone with access already belongs there.

To cover tracks, traffic generated during the attacks was routed through a network of IP addresses spanning more than ten countries.

The developer’s response and South Korea’s reaction

Li Puhua has distanced himself from the hacking activities, emphasizing that ARTEX was created for authorized security testing purposes. Converting the project to closed-source was his response to the alleged misuse, effectively cutting off public access to a tool that had become a liability.

South Korean authorities moved quickly. President Lee Jae Myung publicly addressed the incidents, stating the need for enhanced cybersecurity measures, particularly given the growing threats posed by advanced AI technology in financial sectors. An investigation is underway, though the cross-border nature of the attacks and the routing obfuscation will likely complicate enforcement efforts.

The scale of exposed data, while not catastrophic by the standards of history’s largest breaches, is notably sensitive. Annual income figures and loan limits are the kind of information that enables highly targeted financial fraud.

The bigger picture for AI and cybersecurity

One person, allegedly a single 26-year-old, reportedly breached nearly a dozen financial firms in the span of roughly two weeks.

The ARTEX case also raises uncomfortable questions about the responsibility of foundation model providers. Claude Code was reportedly used alongside the penetration testing agent, raising the issue of whether AI companies need more robust safeguards against their tools being integrated into attack chains. Anthropic has not publicly commented on the incidents as of this writing.

For financial institutions in South Korea and beyond, the immediate priority is assessing whether their internal systems can withstand the kind of AI-augmented probing that ARTEX enabled. The attackers targeted internal infrastructure rather than customer-facing applications, suggesting that perimeter defenses alone are no longer sufficient.