ARTEX: agentic AI pentest tool used against South Korean banks

Between late September and early October 2026, an attacker breached several South Korean financial organisations and stole data from them, relying almost entirely on artificial intelligence tools. CrowdStrike reconstructed the campaign in a report published on 7 October, later covered by Infosecurity Magazine, Security Affairs and The Hacker News. The main tool was ARTEX, a recently released open-source “agentic” penetration testing framework developed in China, which hands control of the test to a language model.

What makes the case stand out is how it came to light: the attacker left open directories on their own servers, holding session histories, configuration files and the AI assistant’s memory files. Researchers could follow the operator’s working method prompt by prompt.

What happened

• Targets: South Korean financial organisations. According to the sources, at one bank the attacker got into a loan progress inquiry service used by financial brokers; at another, they compromised an employee mobile work-support system. The total number of affected organisations has not been confirmed.
• Exfiltration and sale: data was stolen and, according to CrowdStrike, the attacker asked Claude, Anthropic’s model, where Korean breach data is usually sold and how to find Korean Telegram groups that trade it.
• Attribution: CrowdStrike does not link the activity to a known group. It assesses with moderate confidence that the operator is Chinese-speaking and financially motivated.
• Authorities’ response: on 6 October South Korea’s Financial Services Commission issued an alert to customers of the affected companies, urging them to watch out for phishing attempts and loan scams.
• The tool: according to The Hacker News, as reported by Security Affairs, the ARTEX developer has since closed the source and stopped updates, saying the project was built for learning and research and has nothing to do with the attacks. Copies already in circulation remain available.

Technical details

The sources do not mention CVEs or the specific vulnerabilities exploited: they only state that ARTEX was used to discover vulnerabilities and compromise specific services at the victims.

Infrastructure. The attacks were linked through overlapping IP addresses and the string “ARTEX” in HTML files on an attacker-controlled server. The sources describe two servers:

• 38.244.50[.]120, which hosted the ARTEX instance likely used against the banks and an open directory with a .claude/CLAUDE.md file reachable on port 18899. The file contained a Chinese-language prompt telling the model how to conduct the penetration test;
• a Hong Kong address, referenced in that file and assessed as the main infrastructure: more open directories with Claude Code session histories, ARTEX configurations and memory files.

Models used. The ARTEX instance used DeepSeek v4.1-flash as its primary model, probably accessed through the reseller xcai[.]pro; for additional Claude Code sessions the operator also used GLM-5.3 from Zhipu AI and Grok 4.6.

Proxies. The sessions show nine proxy addresses used during the ARTEX-related activity:

• 101.53.80[.]20
• 205.214.59[.]31
• 124.155.252[.]63
• 154.201.79[.]246
• 23.248.249[.]90
• 23.158.220[.]98
• 103.248.148[.]84
• 203.160.133[.]172
• 209.209.85[.]38

The sources do not publish a MITRE ATT&CK mapping or file hashes.

According to CrowdStrike, the case shows how AI tools allow a financially motivated actor to carry out multiple intrusions in a short time, and adversaries will likely keep experimenting with them to raise the tempo and capabilities of their operations.

Impact for European and Italian companies

The known victims are in South Korea, but the attack model has no borders. A tool like ARTEX lowers the barrier to entry: a single operator can probe and attack several targets in parallel, and the time between finding a weakness and exploiting it gets shorter.

The services hit are telling: a portal for brokers and a mobile app for employees, in other words peripheral applications, often exposed to the internet and less closely watched than core systems. For European banks, insurers and financial firms subject to DORA, and for NIS2 entities, a breach of this kind triggers incident reporting obligations. After the breach, as the Financial Services Commission alert shows, the risk shifts to customers, who become targets of phishing and scams built on the stolen data.

What to do right away

1. Search for the published indicators in firewall, proxy and web server logs from recent months: 38.244.50[.]120 (including port 18899) and the nine proxy addresses. IP addresses can be reassigned: a match must be analysed in context, not treated as proof on its own.
2. Inventory your exposed applications, including secondary ones: partner and broker portals, employee mobile apps, inquiry services. These are the kind of targets mentioned in the report.
3. Shrink the vulnerability window: scan exposed services regularly and patch quickly, because an automated attacker moves fast.
4. Detect automated traffic: bursts of requests, enumeration of paths and parameters, repeated attempts against the same endpoints. Apply rate limits and strong authentication on third-party-facing services.
5. Watch for exfiltration: unusually broad queries on customer data, bulk exports, anomalous outbound traffic from application servers.
6. Prepare customer communications: if a breach occurs, warn customers promptly about phishing and scam risks, as the South Korean authority did.

How ARXDOME helps

ARX Expose shows what an attacker sees from outside, vulnerable services included, so a forgotten secondary portal surfaces before someone else finds it. ARX Horizon enriches events with threat lists, and ARX Sentry brings server and firewall logs into the SOC and NOC pages, where you can look for connections to suspicious addresses. ARX Shield keeps track of the open vulnerabilities on monitored machines.

Last updated: 9 October 2026