Microsoft Security ties Cambodia-based Storm-3358 to ARTEX financial-sector attacks

Microsoft Security reported on Nov. 7, 2026 that Storm-3358, a Cambodia-based intrusion cluster tracked by Microsoft Threat Intelligence, expanded in October 2026 into the financial sector using ARTEX, an agentic AI offensive framework. The company said the cluster focuses mainly on financial organizations across Southeast Asia, though it also hits a limited number of major financial firms around the world.

Storm-3358 pairs ARTEX with conventional automated tools when it scans internet-connected systems, the company reported, and ARTEX agents use netcat and curl to download remote access trojans, malware that lets attackers control compromised machines, to those systems. The company also said ARTEX’s default tools include Nmap, curl, wget, Python HTTP clients, Norma SDK WebFetch, and headless Chromium through Playwright.

The cluster takes advantage of anonymous file upload vulnerabilities and server-side request forgery issues, the company said, a flaw that lets an attacker force a server to make requests to internal systems. Attacks linked to ARTEX involved scanning for web application vulnerabilities and exploiting them, flaws in software that serves web pages, along with web shell use, which places scripts on compromised servers so attackers can keep access. The company observed Storm-3358 installing WISP, vshell, and AdaptixC2, open-source remote access trojans used for persistence and lateral movement.

Attackers in ARTEX-related attacks drew credentials from public sources and from breached systems, the company said, then relied on compromised cloud identities to keep access and steal data. ARTEX agents can run brute-force password attacks, which try many passwords against one account, and password spraying, which tries a few passwords across many accounts, to guess login credentials.

The agents can also perform API enumeration and abuse to locate and misuse programming interfaces, the company said, and dump databases to copy their contents. Agentic tools such as ARTEX can use stolen credentials and compromised systems at machine speed to move through networks, raise privileges, and remove data.

The activity shows criminal actors using agentic AI penetration testing frameworks to speed up every attack stage, the company said, citing ARTEX as one such framework. An agentic AI framework lets software agents plan and carry out steps autonomously instead of following fixed scripts. The company said attackers routinely send traffic through proxies and virtual private networks to change source IP addresses and avoid static detections, security tools that flag fixed patterns such as known addresses.

Agentic AI frameworks issue many different requests quickly when scanning a target, the company said, then try different versions and move to later steps at machine speed. Network traffic from the tools may show distinctive traits, such as default user agent strings, identifiers that reveal the software sending the request. Halting traffic that carries those traits would not stop an AI agent attack, since the agent quickly alters them, although spotting the traits can warn defenders about the activity. Defenders can use indicators of compromise, warning signs such as IP addresses, that the company supplies to spot related activity.

Decoys, tarpits, fake wins, and honeypots can pull autonomous AI attacks, attacks carried out by software acting on its own, away from real targets, the company said. Microsoft Sentinel Deception, the company’s deception service, offers honeypots and decoy surfaces that present worthless targets to agents. Tarpits consume an agent’s time and budget, while fake wins can make an agent halt on a false success. The company said its tests showed the controls reduced attack traffic against real assets and produced high-confidence attribution-quality signal.

The company integrates research on agentic frameworks into Microsoft Entra ID Protection, Microsoft Defender for Endpoint, the Microsoft Defender portal, and Security Dashboard for AI. Security Copilot runs within Microsoft Defender and the company sells it separately as well, the company said, and its AI agents cover Threat Intelligence Briefing, Phishing Triage, Threat Hunting, and Dynamic Threat Detection. Customers with Defender XDR can run advanced hunting queries in the Defender portal or in Microsoft Sentinel to look for related activity.

Defenders can use the Microsoft Sentinel IOC query to check IP addresses across data sources, the company said, looking back 30 days. The company supplied sample IOC IP addresses including 199.19.95.66, 199.19.95.67, 199.19.95.68, 91.92.240.238, 85.11.168.45, 178.15.54.129, 31.77.229.35, and 85.11.167.52.