South Korean Financial Firms Hit by AI Pentest Tool Breach Exfiltrating 67k Records

South Korean financial institutions suffered a data breach after a threat actor abused an artificial intelligence penetration testing framework to steal personal records. According to Zscaler ThreatLabz, the campaign targeted peripheral internet-facing web applications, including customer-facing inquiry portals and employee-facing mobile application backends. The malicious code, identified as ARTEX, is an open-source agentic framework developed in China that uses large language models to automate reconnaissance and exploitation tasks. This tool lowers the technical barrier for conducting complex intrusion campaigns by orchestrating multiple AI backends.

How it unfolded

• The campaign was first observed in late September 2026.
• The threat actor deployed ARTEX to automate the scanning and exploitation of web applications.
• The attack resulted in the exfiltration of personally identifiable information belonging to approximately 67,000 individuals.
• Zscaler ThreatLabz assessed with moderate confidence that a financially motivated, Chinese-speaking actor was responsible.

Who is affected

Multiple South Korean financial firms were impacted by this exploitation campaign. The breach resulted in the loss of personally identifiable information for roughly 67,000 individuals. The attackers targeted specific external entry points. They exploited customer-facing inquiry portals and backends for employee mobile applications. These systems are described in the source as peripheral internet-facing web applications. The source does not specify whether core banking systems were accessed or if these peripheral systems were chosen due to weaker controls.

The fix

No patch or update has been confirmed yet for this specific campaign. The source material does not mention a vendor release addressing the vulnerabilities exploited. Organizations are advised to monitor for further guidance. Zscaler ThreatLabz provided recommendations to strengthen defenses against AI-assisted attacks, including reducing application exposure and establishing visibility into internet-facing assets. Teams should also consider deploying deception controls to detect suspicious reconnaissance attempts.

What analysts watch for

Analysts typically examine logs for sequential lookups and unusually high record access rates on partner or employee portals. They check whether authorization checks are enforced server-side rather than relying on client-side restrictions. Teams look for signs of automated reconnaissance, such as rapid scanning patterns consistent with agentic frameworks. Investigators also review configurations of open-source penetration testing tools to ensure they are isolated from production environments. Monitoring for prompt injection attempts and model drift in internal AI systems is also standard practice.

Background: Threat actors

A threat actor is any individual or group that intentionally seeks to compromise information systems. They may be motivated by money, ideology, or espionage. Understanding their methods helps you defend against specific techniques rather than abstract fears.

Read the full guide: Threat Actors Definition: Who They Are and How They Operate

What to do and how to stay safe: ARTEX

• Reduce external application exposure by moving services behind a Zero Trust access solution.
• Establish complete visibility into internet-facing and internal assets to assess their risks.
• Alert on sequential lookups and high record access volumes on portals and apply rate limits.
• Isolate open-source pentesting harnesses from production and avoid storing credentials in them.

Step-by-step guide: Incident Response Plans: Real Benefits and Hidden Costs

General security guidance from the Malware Brief newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

What tool did the attacker use in this campaign?

The attacker used ARTEX, an open-source agentic penetration testing framework that uses large language models.

How many records were stolen?

Approximately 67,000 records of personally identifiable information were exfiltrated.

Who is behind the attack?

Zscaler ThreatLabz assesses with moderate confidence that a financially motivated, Chinese-speaking actor is responsible.