Threat Actor Uses AI Framework to Steal Data From South Korean Banks
A Chinese-speaking group used an agentic AI tool to breach South Korean financial firms, stealing records of 67,000 people in late 2026.
Key points
- Zscaler ThreatLabz identified a campaign using the ARTEX framework to target South Korean financial institutions.
- The attack resulted in the theft of personally identifiable information belonging to approximately 67,000 individuals.
- Analysts attribute the breach to a financially motivated, Chinese-speaking threat actor operating in late September 2026.
Zscaler ThreatLabz reported that a threat actor abused the ARTEX penetration testing framework to compromise multiple South Korean financial institutions in late September 2026. The campaign resulted in the exfiltration of personally identifiable information belonging to approximately 67,000 individuals, according to the security firm's analysis.
The attackers exploited peripheral internet-facing web applications, including customer-facing inquiry portals and employee-facing mobile application backends. Zscaler ThreatLabz assesses with moderate confidence that the campaign was conducted by a financially motivated, Chinese-speaking threat actor. The group leveraged the open-source tool to automate reconnaissance and exploitation tasks during the intrusion.
The AI-Powered Attack Tool
ARTEX is an agentic penetration testing framework originally released on GitHub by the developer Autumn-27. The tool orchestrates multiple large language model backends to autonomously perform reconnaissance, vulnerability identification, and exploitation. This automation significantly lowers the technical barrier for conducting complex intrusion campaigns, according to Zscaler.
The threat actor configured ARTEX to use specific AI backends to support its operations. The framework allows for automated execution of tasks that typically require manual effort from skilled penetration testers. This capability enabled the attackers to efficiently identify and exploit weaknesses in the targeted financial firms' external applications.
Impact on Financial Sectors
The campaign affected multiple financial firms in South Korea, leading to significant data loss. The stolen data included personally identifiable information for roughly 67,000 individuals. The attackers focused on peripheral systems rather than core banking infrastructure, targeting inquiry portals and mobile app backends.
Zscaler ThreatLabz noted that the use of AI in this campaign highlights the evolving nature of cyber threats. The ability to automate complex tasks allows threat actors to scale their operations and target a wider range of vulnerabilities. This incident underscores the risk posed by open-source security tools when misused by malicious actors.
What should you do?
Zscaler recommends reducing application exposure by moving applications behind Zero Trust access solutions. Organizations should establish complete visibility into internet-facing and internal assets, including AI assets, to assess their exposure and vulnerabilities. Deploying deception controls can help detect suspicious reconnaissance and access attempts through decoy assets and accounts.
Additionally, Zscaler advises prioritizing Zero Trust everywhere architecture to enforce least-privilege user-to-application access. This approach limits lateral propagation and reduces the blast radius from AI-driven attacks. Organizations should also use AI red teaming and guardrails to assess susceptibility to prompt injection and unsafe outputs.
Background: Large language models
Large language models are a specific type of neural network trained on text. Generative AI is the broader umbrella term for any system that creates new content, including images and code. Not all generative AI uses large language models, but most modern text generators do.
What to do and how to stay safe: Zscaler
- Reduce external application exposure by implementing Zero Trust access controls to limit unauthorized access to internet-facing services.
- Establish complete visibility into all internet-facing and internal assets, including AI systems, to identify and prioritize remediation of vulnerabilities.
- Deploy deception controls such as decoy assets and accounts to detect suspicious reconnaissance activities and automatically contain potential threats.
- Enforce server-side authorization for access to sensitive records and limit responses to only the fields required for specific operations.
Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is ARTEX and how was it used in this attack?
ARTEX is an open-source agentic penetration testing framework that uses large language models to automate reconnaissance and exploitation tasks. The threat actor used it to target South Korean financial institutions.
How many people were affected by this data breach?
Approximately 67,000 individuals had their personally identifiable information exfiltrated during the campaign targeting South Korean financial firms.
Who is responsible for this cyberattack?
Zscaler ThreatLabz assesses with moderate confidence that a financially motivated, Chinese-speaking threat actor is responsible for the campaign.
