Storm-3358 Turns ARTEX AI Hacking Framework on Banks

Storm-3358 turns ARTEX, an AI hacking framework, on banks. That's the finding in a Nov. 6, 2026 post from Microsoft Threat Intelligence, which says the financially motivated group began using the tool against financial-sector organizations in October.

The group isn't new to Microsoft's tracking, and the tool isn't new either. What's new is who is holding it: the same framework that Microsoft tied to a China-state actor two weeks ago is now in the hands of a cybercrime cluster.

Microsoft Threat Intelligence's post on X, Nov. 7, 2026, announcing the Storm-3358 findings and linking to the blog (@MsftSecIntel)

View on X

A Cambodia-based cluster with a bank list

Microsoft describes Storm-3358 as a Cambodia-based intrusion cluster. It mostly targets financial organizations in Southeast Asia, plus a smaller number of large financial organizations elsewhere. The post doesn't name any victim or say how many were hit.

According to Microsoft, the group uses ARTEX alongside conventional automated tools to scan internet-facing systems. It then exploits flaws that include anonymous file upload and server-side request forgery (SSRF), where a server is tricked into making requests on an attacker's behalf.

What ARTEX is, and who else uses it

In its Oct. 21 post on the China-aligned actor Storm-0940, Microsoft calls ARTEX an open-source penetration testing framework that can run on any AI model. It is a multi-agent system written in Go with a Next.js frontend, and its agents can carry out a test from reconnaissance through data exfiltration with little human input.

That post said Storm-0940 used it from September against hundreds of organizations, and that in several cases the time from first targeting to privileged cloud access was under three hours. Then, on Nov. 5, Microsoft described threat actors using ARTEX to hunt for exposed secrets on GitHub, including one case that went from a secrets scan to Azure Key Vault exfiltration in five minutes.

Our count: that's three ARTEX posts in 16 days, covering a state actor, a credential-hunting campaign, and now a criminal group. Microsoft's own wording is that this is "an ongoing trend."

From web flaw to remote access

Once inside, Microsoft says Storm-3358 installs open-source remote access trojans (RATs): WISP, vshell and AdaptixC2, used for persistence and lateral movement. Microsoft observed ARTEX agents using netcat and curl to download them to compromised systems.

The post lists other agent behavior seen in recent ARTEX attacks: web shells on compromised servers, credentials taken from public sources, brute-force and password-spraying attempts, API enumeration, database dumping, and compromised cloud identities for persistence and exfiltration. Microsoft doesn't say which of these Storm-3358 itself used, apart from the scanning, exploits and RATs above, so don't read the whole list as this group's playbook.

The IP table has a typo-sized hole

Microsoft published eight sample indicators. Four are ARTEX servers, one is a C2 server and three are WISP C2 nodes, all marked active as of Nov. 6.

Here's the catch. The Microsoft Sentinel hunting query in the same post lists 178.15.54.129, not 178.16.54.129. One of them is a typo, and the table is the version labeled as an ARTEX server. We checked the public RIPE registry: 178.16.54.129 sits in a Turkish hosting block (TR-MGN), the same registry name as another listed ARTEX server, 91.92.240.238. 178.15.54.129 belongs to a Vodafone Germany consumer range. That points to the table being right and the query being wrong, though Microsoft hasn't confirmed it. If you paste the query as published, it may never match traffic to the server first seen Oct. 19, the oldest one on the list.

Microsoft also warns that attackers routinely hide behind proxies and VPNs, so these addresses are samples, not a blocklist that finishes the job.

Spotting a machine-speed attacker

Microsoft's detection advice is behavioral. An agent scanning a target sends a wide variety of requests in a short time, retries variations of the same technique, and moves to next-stage actions at machine speed. By default, ARTEX uses off-the-shelf tools including Nmap, curl, wget, Python HTTP clients and headless Chromium through Playwright, which can leave default user-agent strings in your logs.

Blocking those strings won't stop an agent, Microsoft says, because it will change them quickly. The value is as an early alert.

The more unusual advice is deception. Microsoft recommends honeypots, decoys, tarpits that burn an agent's time and budget, and "fake wins" that convince it the objective is done. In Microsoft's testing, these reduced attack traffic reaching real assets and produced high-confidence signals. That's Microsoft's claim about its own tests; the post doesn't publish figures. For a defender, it's the same logic as CrowdStrike's push to buy agent-security startups: the attacker is now software, so the target should include software-shaped traps.

Four fast answers

Who is Storm-3358? A Cambodia-based, financially motivated intrusion cluster that Microsoft says mostly targets financial firms in Southeast Asia.

What did it exploit? Anonymous file upload and SSRF flaws on internet-exposed systems, per Microsoft. No specific products or CVEs are named.

Which malware did it drop? The open-source RATs WISP, vshell and AdaptixC2.

Unconfirmed: how many organizations were breached, whether any money was stolen, and which exposed systems were the entry points.

Check these before Monday

The practical list is short. Search your logs for the eight addresses above, using 178.16.54.129 (and 178.15.54.129 only if you want to rule out the query's version). Review any public-facing upload endpoint that accepts files without signing in, and any feature that fetches a URL on a user's behalf, since those are the two flaw types named. And ask whether your alerting can handle a burst of varied requests from one source at 3 a.m., because that pattern is the one Microsoft says gives these agents away.

Our bet is that Microsoft's fix will keep the .16 address; watch whether the post gets a correction note.

Source: Microsoft Security Blog

artex storm 3358 microsoft cybersecurity agentic ai

AI

About the author

AI Briefing Daily Editorial. The editorial desk of AI Briefing Daily, an independent publication covering artificial intelligence. Every story is researched against primary sources, checked by hand and reviewed by a person before publication.