A threat actor paired the ARTEX agentic pentest tool with Claude Code to breach South Korean financial organizations
2026-artex-claude-code-korea-bank-breach · 2026-10-08
CrowdStrike Intelligence reported a late-September to early-October 2026 campaign in which a financially motivated operator used ARTEX, a China-developed open-source agentic penetration-testing tool, alongside Claude Code and other LLM backends to find and exploit vulnerabilities in multiple South Korean financial organizations, resulting in data exfiltration. It was uncovered when the operator left open directories exposing Claude Code session histories and ARTEX configuration files.
System
- Framework
- ARTEX agentic penetration-testing tool with Claude Code
- Models
- claude, deepseek, glm, grok
- Tools
- artex, claude-code, pentesting-utilities
- Vendor
- unattributed operator (suspected China-based, financially motivated)
- Autonomy
- supervised-autonomous
Classification
- Primary class
- autonomous-misuse
- Chain
- autonomous-misuse/cyber-ops → data-exfiltration/via-tool
- Attack vector
- direct-user
- Causation
- entity: human · intentionality: intentional · timing: post-deployment
Trigger
The operator ran ARTEX, a recently released open-source agentic pentesting tool developed in China, from attacker-controlled infrastructure, and used Claude Code sessions for vulnerability research and attack planning. ARTEX was driven by LLM backends (DeepSeek v4.1-flash as primary, plus GLM-5.3 and Grok, accessed via a likely API proxy) to identify and exploit exposed banking services, including a loan progress inquiry service and an employee mobile work-support system.
Root cause
An open-source agentic pentesting tool combined with capable LLM backends and the Claude Code interface turned a single operator's intent into an accelerated attack pipeline for reconnaissance, vulnerability identification, and exploitation, enabling multiple intrusions within a short time span.
Contributing factors
- An open-source agentic pentesting tool (ARTEX) packaged offensive automation around LLM backends.
- Claude Code sessions were used for vulnerability research and attack planning.
- Internet-exposed banking services presented exploitable attack surface.
Detection
Discovered by CrowdStrike Intelligence, which uncovered threat actor-controlled open directories exposing Claude Code session histories, Claude memory files, and ARTEX configuration files; disclosed on 2026-10-08. CrowdStrike assessed with moderate confidence that the operator was a likely Chinese-speaking, financially motivated actor, based on the use of the China-developed ARTEX tool and observed Chinese-language prompts, and recovered possible identifying details that it stated were not reliable enough to confirm the actor's identity.
Recovery
Reported publicly by CrowdStrike Intelligence; the ARTEX tool was subsequently taken closed-source after disclosure. Mitigations centre on patching exposed financial-sector services, monitoring for agentic-attack behaviour, and abuse detection against offensive orchestration of AI models.
Prevention
Reduce exposure and patch internet-facing financial services; monitor for machine-speed vulnerability research and exploitation patterns; constrain offensive orchestration of agentic tooling and LLM backends via usage policy and abuse detection.
Blast radius
- Data
- Exploitation of exposed banking services across multiple South Korean financial organizations, including a loan progress inquiry service and an employee mobile work-support system; reporting indicates exposure of client personal data and payment card information. Exfiltrated data volumes were not quantified. pii
- User harm
- Compromise of financial-sector systems and exposure of customer data across multiple institutions. economic legal-regulatory
- Scope
- multiple South Korean financial organizations
- Reversibility
- irreversible
References
- OWASP LLM
- LLM02
- MITRE ATLAS
- AML.T0024
- Related
- 2025-gtg1002-ai-orchestrated-espionage 2026-deepseek-hermes-autonomous-campaign 2026-taiwan-nuclear-agency-autonomous-attack
- Tags
- agentic-attack cyber-ops financial-sector artex claude-code pentesting-tool
Sources
Cite this incident
Permalink: https://swarmproof.github.io/agent-postmortems/2026-artex-claude-code-korea-bank-breach/
@misc{2026-artex-claude-code-korea-bank-breach,
title = {A threat actor paired the ARTEX agentic pentest tool with Claude Code to breach South Korean financial organizations},
year = {2026},
howpublished = {agent-postmortems},
url = {https://swarmproof.github.io/agent-postmortems/2026-artex-claude-code-korea-bank-breach/}
}