✅ Report Title: Initial Analysis of ARTEX, an AI Pentesting Framework Abused in an Attack on South Korea's Financial Sector

S2W TALON confirmed that ARTEX, an open-source autonomous penetration testing tool, was used in a recent attack targeting South Korea's financial sector, and conducted an analysis of the tool.

📌 What is ARTEX?

ARTEX is an open-source autonomous penetration testing system that utilizes a Large Language Model (LLM) and multiple agents. The tool is designed so that when a user specifies a goal, the agent plans the inspection direction based on it and invokes the actual tool to perform asset discovery and vulnerability verification.

📌 Key Features of ARTEX

• (Workflow)
  • When an authenticated user inputs a task goal and scope, the goal decomposition agent extracts the goal, constraints, and scope, and the Planner generates individual tasks (intents) to be executed.
  • The Worker is assigned an intent, executes a tool, and saves the results, while the Planner plans subsequent tasks based on these results. This process repeats iteratively, and the task concludes once all goals are completed.
• (Agent Roles) ARTEX consists of Planners and Workers responsible for task planning and execution, as well as role-specific LLM agents responsible for handling user requests, general conversation, report generation, and vulnerability re-verification.
• (Tool Execution) Each agent performs tasks using basic SDK tools and ARTEX-specific tools, depending on its role and configuration.
• (Execution Control) Applies rules to execution requests from tools subject to review, determining whether to allow, block, or require user approval.
• (Asset Scope Control) Manages asset registration and the creation of intents connected to assets by applying global and task-specific asset allow and block rules.
• (Traffic Recording) When enabled, saves HTTP and HTTPS requests and responses passing through the proxy, and provides search, detailed lookup, and deletion functions.
• (Asset Management) Manages assets and target scope per company.
• (Finding Management & Reporting) Provides a function to manage information such as vulnerability type, severity, summary, and PoC.
• (Tool & Skill Integration) ARTEX enables the expansion of agent functions through user-defined tools, MCP servers, and Skill integration.
• (LLM Management) Uses APIs to connect to LLM models, with model settings configurable per task and per agent.

✅ Threat Detection Recommendations and Mitigation Measures:

• (Attack Surface Reduction) Vulnerabilities in web services and APIs exposed to the internet must be regularly checked and security updates applied.
• (Web Attack Detection & Blocking) Detect and block attempts to exploit known vulnerabilities using a WAF and IPS, and review the company's web access logs for traces of ARTEX usage linked to attack activity.
• (Execution Control) The use of unnecessary scripts and executable files on servers and endpoints must be restricted, and application execution control and EDR must be applied.
• (Network Access Control) External public servers and key internal systems must be separated, and communication must be restricted to only the targets and ports necessary for business operations.
• (Infrastructure Monitoring) Monitor externally exposed ARTEX infrastructure using the HTML title of the ARTEX management page and the content hash of its login page.
• (Sensitive Data Exposure) If ARTEX instances or related record data are obtained, check whether traffic recording is enabled and its storage scope, and investigate whether the victim company's authentication information, session tokens, and business data are included.