After Korean bank hack, ARTEX developer takes the AI agent closed-source

The Chinese developer of ARTEX, an open-source AI penetration-testing tool, has converted the project to closed source, according to Reuters. The move follows CrowdStrike's disclosure that the tool was used in attacks on South Korean financial institutions.

In a blog published on October 7, CrowdStrike said a threat actor combined ARTEX with large language models to target South Korean financial firms between late September and early October 2026, successfully exfiltrating data.

ARTEX is an “autonomous penetration testing” system driven by multiple AI agents that can automatically find and exploit vulnerabilities. CrowdStrike said the attacker primarily used it to discover and compromise specific services at victim organizations.

Investigators also reconstructed the attacker's toolchain. The ARTEX instance in question used a DeepSeek model as its main LLM backend, supplemented by other models, and the attacker asked Claude where stolen Korean data is typically sold.

Several South Korean financial firms were affected, with data exfiltrated. CrowdStrike assessed with moderate confidence that the attacker is a Chinese-speaking, financially motivated operator, though the activity has not been attributed to any known group.

Responding to the attention, ARTEX developer Autumn-27 denied any connection to the malicious attacks, stressing that the tool was designed for authorized security testing and risk assessment of assets, and that such misuse runs against its original purpose.

As a result, the developer said the project would no longer be open-sourced and that updates, releases and maintenance support would end — meaning ARTEX will no longer ship as publicly available source code.

The episode is widely seen as a landmark case for AI-agent security risk: once a dual-use open-source tool is weaponized, developers struggle to control its spread and can only cut their exposure by tightening distribution.

What to watch next is how regulators and security vendors treat “AI penetration testing” tools, and how the open-source community's debate over the responsibility boundaries of security tooling evolves.