# TAG-217 Uses Claude Code, DeepSeek, Grok and ARTEX in AI-Assisted Attacks on South Korean Financial Institutions

> Recorded Future's Insikt Group tracks TAG-217, a suspected Chinese-speaking actor (not confirmed as state-backed), using AI tools (Claude Code, DeepSeek, Grok, Zhipu AI GLM) and the Chinese-language agentic penetration-testing framework ARTEX against at least seven South Korean financial institutions. Personal information of roughly 68,000 people was reported exposed.

- **Published:** 2026-10-19T06:43:27Z
- **Last reviewed:** 2026-10-21T21:08:49.782Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-3860
- **ID:** TL-2026-3860
- **Severity:** HIGH
- **Category:** THREAT_INTEL
- **Status:** ACTIVE
- **Actor:** TAG-217
- **Detections:** 9 · **IOCs:** 40 (full data via the Threadlinqs MCP server — Purple tier)

## Description

In October 2026 Recorded Future's Insikt Group (report of 2026-10-16) described TAG-217, a suspected Chinese speaker who is not confirmed as a state-backed group, using commercial and open AI services together with ARTEX, an open-source Chinese-language framework for automated security testing, against South Korean financial organizations. Exposed files on actor-controlled infrastructure revealed the preparation workflow: infrastructure testing and scanning of internet-facing systems from 2026-08-17 to 2026-08-22, assembly of cryptocurrency-targeting toolkits on 2026-08-22, creation of an iPhone/Android mobile testing environment on 2026-09-05, and operational AI tooling plus reconnaissance infrastructure by 2026-09-26. Recorded Future states that the exposed files showed preparation but did not confirm successful bank intrusions, and that deployment of the mobile and crypto tools was unconfirmed.

Observed tradecraft includes routing traffic through residential proxies, YiiMP-targeting scripts that try default passwords and send Telegram alerts, cryptocurrency wallet-stealing tools, Linux privilege-escalation exploits, and mobile device testing. AI services named across reporting are Claude Code, DeepSeek, Grok and Zhipu AI. Recorded Future states that basic operational security mistakes left a trail.

CrowdStrike Intelligence independently reported (without using the TAG-217 name) infrastructure tied to a campaign against South Korean financial organizations, active from late September to early October 2026, in which data was exfiltrated. It assessed with moderate confidence that the actor is a financially motivated Chinese speaker, based on use of ARTEX and Chinese-language prompts. Open directories held Claude Code session histories, ARTEX configuration files and Claude memory files. These showed a two-server architecture (an ARTEX/Claude Code host at 38.244.50.120 plus separate attack infrastructure), nine proxy IPs, and the LLM API proxy xcai.pro, likely a DeepSeek API reseller. The LLM backends were DeepSeek v4.1-flash (primary), GLM-5.3 and Grok 4.6. The session records indicate the actor had no specific monetization plan and asked Claude to suggest Korea-focused Telegram data-sale groups. Whether CrowdStrike's activity and TAG-217 are the same actor is not established in the sources reviewed.

South Korean press and AhnLab reporting (2026-10-06) cites seven financial firms and about 68,000 people's data exposed, and says AhnLab's Security Intelligence Center observed ARTEX instances on roughly 600 IPs worldwide. That count does not establish that each server was used in an attack or run by the same group. South Korea's National Police Agency is investigating. Banks reported as affected include Shinhan Bank, KB Kookmin Bank, Hana Bank, BNK Busan Bank and Yegaram Savings Bank; exfiltrated data included personal and credit-card information, and outages were reported. A trace of ARTEX on an attack server does not by itself confirm that an AI agent conducted the intrusion. The ARTEX developer reportedly made the project closed-source after the real-world attacks, though derivatives remain available. Recorded Future states that detailed campaign IoCs are in its customer intelligence-platform note, so the public IOC set here is limited to what CrowdStrike published.

## MITRE ATT&CK

- T1595.001 Scanning IP Blocks
- T1595.002 Vulnerability Scanning
- T1583.003 Virtual Private Server
- T1588.007 Artificial Intelligence
- T1588.002 Tool
- T1059 Command and Scripting Interpreter
- T1110.001 Password Guessing
- T1090 Proxy
- T1190 Exploit Public-Facing Application
- T1590 Gather Victim Network Information
- T1059.004 Command and Scripting Interpreter
- T1046 Network Service Discovery
- T1119 Automated Collection
- T1090.002 Proxy
- T1572 Protocol Tunneling
- T1567.002 Exfiltration Over Web Service

## Sources

- [Hackers Use Claude, DeepSeek and Grok in AI-Assisted Attacks on South Korean Banks](https://gbhackers.com/ai-assisted-attacks/)
- [Unknown Threat Actor Uses AI-Driven ARTEX to Target South Korean Finance (CrowdStrike)](https://www.crowdstrike.com/en-us/blog/unknown-threat-actor-uses-artex-to-target-south-korean-finance/)
- [Hacker used ARTEX AI and Claude agents to target South Korean banks (BleepingComputer)](https://www.bleepingcomputer.com/news/security/hacker-used-artex-ai-and-claude-agents-to-target-south-korean-banks/)
- [AhnLab finds ARTEX on 600 IPs as police probe South Korean bank breaches](https://runtimewire.com/article/south-korea-bank-breaches-artex-ai-probe)
- [AI-powered attacks on South Korean financial sectors linked to Chinese DDoS group: report](https://asianews.network/?p=307184)
- [ARTEX GitHub repository (as cited by RuntimeWire)](https://github.com/Autumn-27/ARTEX)

## Full data

Detection queries (Splunk SPL / Microsoft KQL / Sigma) and IOC values require the Threadlinqs MCP server (Purple tier): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-3860