Microsoft says a Cambodia-based crime group has turned the AI hacking framework ARTEX on the financial sector, using autonomous agents to scan banks’ internet-facing systems, exploit web flaws and plant remote-access malware. The group, tracked as Storm-3358, began the campaign in October, and Microsoft says it mainly targets financial organisations in Southeast Asia, with a smaller number of large financial firms elsewhere also in scope.
From stolen secrets to bank networks
The new report is a change of target, not a change of tool. Earlier this week, Microsoft’s Threat Intelligence team reported that ARTEX agents were hunting exposed secrets in public code and reaching cloud vaults within minutes. Storm-3358 uses the same framework differently, aiming it at live financial systems rather than at code repositories.
Microsoft says the group pairs ARTEX with conventional automated tools to scan systems connected to the public internet. The agents exploit flaws including anonymous file upload and server-side request forgery, then install open-source remote access trojans known as WISP, vshell and AdaptixC2 to stay inside the network and move sideways.
Machine speed, from scan to theft
The speed is the point. Microsoft says ARTEX agents can brute-force passwords, spray them across accounts, enumerate and abuse programming interfaces, and dump databases, all without waiting for a human operator. Once inside, they use compromised credentials and cloud identities to move laterally, escalate privileges and exfiltrate data.
The agents also fetch their own malware. Microsoft observed ARTEX agents using ordinary utilities such as netcat and curl to download the remote access trojans onto compromised systems. The company has published indicators of compromise, and several of the listed addresses were still active on 6 November, when the report was published.
The defence Microsoft recommends
Blocking the tools will not work, Microsoft says, because an agent that hits a blocked signature simply changes it and tries again. Instead, the company recommends deception: honeypots, decoys, tarpits and fake wins that pull autonomous attacks towards worthless targets. In Microsoft’s testing, these controls reduced the attack traffic reaching real assets.
The framework itself runs on familiar utilities, which gives defenders a signal. ARTEX by default uses Nmap, curl, wget, Python web clients and headless Chromium through Playwright, and the default traffic patterns of those tools can alert defenders. Microsoft also points to its Defender detections and a new Security Dashboard for AI that tracks the risk.
Our Read
This report matters because it changes who should worry about ARTEX. Until now the story was about spies and stolen secrets, which let most companies treat it as someone else’s problem. A financially motivated crew aiming the same framework at banks puts every firm with an application in scope, and the attack needs no zero-day, only an unpatched web flaw.
The timing sharpens the contrast. In the same week, OpenAI introduced a defensive model built to find flaws before attackers do, while Microsoft documents attackers automating the other side of the same work. Defence and offence are now both agent-driven, and the side that adapts faster wins the minutes that matter.
Why It Matters
Banks and financial firms can no longer treat agentic hacking frameworks as a research curiosity or a nation-state problem. The tools are open, the techniques are ordinary web exploits, and the operators now include crews that want money rather than secrets. The window between a flaw appearing and an agent exploiting it is shrinking towards zero.
Background
ARTEX has surfaced repeatedly in recent weeks. Zscaler documented one operator using it across 19 countries, and CrowdStrike described an operator steering agents at more than 200 organisations. Microsoft’s own reports now link the framework to a nation-state supply-chain campaign, the GitHub secret-hunting operation and this financial-sector campaign.
Source: Microsoft Threat Intelligence report and Technobezz report
0 Comments