High 4 Hunt LeadsIOCsStorm-3358WISPvshellAdaptixC2Microsoft Threat Intelligence

The ARTEX AI Attack Kit Goes Criminal, and It’s Aimed at Banks

Microsoft says Cambodia-based Storm-3358 began using the ARTEX agentic AI framework against financial organizations in October, exploiting web app flaws and installing WISP, vshell and AdaptixC2 RATs.

The AI attack framework that helped a Chinese state group break into ConnectWise is now in criminal hands, and it’s pointed at banks. Microsoft Threat Intelligence says a financially motivated group it tracks as Storm-3358, a Cambodia-based intrusion cluster, started using ARTEX against financial organizations in October. Its targets are mostly in Southeast Asia, plus a smaller number of large financial firms worldwide. It’s the third ARTEX report from Microsoft in three weeks, after Storm-0940’s supply-chain attack and agents raiding secrets leaked on GitHub.

For defenders

What Storm-3358 does with it: ARTEX agents, working alongside ordinary automated scanners, sweep internet-facing systems (T1595.002 (Active Scanning: Vulnerability Scanning)), exploit flaws such as anonymous file upload and server-side request forgery (T1190 (Exploit Public-Facing Application)), then pull down open-source remote access trojans with netcat and curl (T1105 (Ingress Tool Transfer)). Microsoft names three: WISP, vshell and AdaptixC2, used for persistence and lateral movement. Microsoft doesn’t name the specific products or CVEs that were exploited.

The agent’s habits: lots of varied requests in a short window, quick retries of the same technique with small changes, and a jump to the next stage at machine speed. ARTEX’s default toolset includes Nmap, curl, wget, Python HTTP clients and headless Chromium via Playwright, which can leave default user agents in your logs. Microsoft’s warning: blocking those strings won’t stop an agent that changes them in seconds, but spotting them tells you one is there. It also says honeypots, tarpits and “fake wins” pulled agents away from real assets in its testing.

Hunting: Microsoft published a Defender XDR query for curl downloading files a host hasn’t fetched before, and a Sentinel query for the IPs below; both are in the Hunt Lead report. Its own query lists one ARTEX server as 178.15.54.129 while its indicator table says 178.16.54[.]129, so search for both.

Network
199.19.95[.]66 · 199.19.95[.]67 · 199.19.95[.]68 (WISP C2 nodes)
91.92.240[.]238 · 85.11.168[.]45 · 178.16.54[.]129 · 31.77.229[.]35 (ARTEX servers)
85.11.167[.]52 (C2 server)

What it means for you

This is a warning for banks, payment firms and anyone running internet-facing web apps, not for individual customers. Nothing here says customer accounts were taken. The bigger shift: tools that used to need a skilled hacker now let a small criminal crew probe and break into servers around the clock, so unpatched web apps get found faster than before.

Do this now

  1. Patch and lock down anything you expose to the internet, especially file-upload features and anything that fetches URLs on a user’s behalf.
  2. Search firewall, proxy and DNS logs for the IPs above, going back to mid-October.
  3. Alert when a web server starts curl, wget or netcat to download a file.
  4. Consider decoy services and fake credentials; automated attackers fall for them far more often than people do.

Source article: ARTEX abuse expands as cybercriminals target financial sector
Publisher: Microsoft Threat Intelligence (Microsoft Security Blog)
Published: Nov. 6, 2026
URL: https://www.microsoft.com/en-us/security/blog/2026/11/06/artex-abuse-expands-as-cybercriminals-target-financial-sector/

Executive Summary

Microsoft Threat Intelligence reports that Storm-3358, a financially motivated, Cambodia-based intrusion cluster that mainly targets financial organizations in Southeast Asia and a smaller number of large financial organizations globally, began using the agentic AI offensive framework ARTEX against the financial sector in October 2026. ARTEX is the same framework Microsoft documented in Storm-0940’s ConnectWise supply-chain compromise and in unattributed campaigns that harvest secrets from public GitHub repositories; Microsoft gives no other vendor name for Storm-3358. In this activity the agents work alongside conventional automated tooling to scan internet-facing systems, exploit weaknesses including anonymous file upload and server-side request forgery, and use netcat and curl to download open-source remote access trojans (WISP, vshell and AdaptixC2) that the group then uses for persistence and lateral movement. Across recent ARTEX intrusions Microsoft also describes web shells on compromised web servers, credential theft from public sources and compromised systems, brute force and password spraying, API enumeration and database dumping, all at machine speed. The objective is financially motivated access to financial institutions; Microsoft does not describe specific victims, losses or exploited CVEs. Microsoft published eight IP indicators (three WISP C2 nodes, four ARTEX servers and one C2 server) active as of Nov. 6, 2026, plus a Defender XDR and a Sentinel hunting query. Hunt leads, false positive expectations, the published queries and indicators are detailed below.

ATT&CK chain: Reconnaissance → Initial Access → Execution → Command and Control → Persistence → Lateral Movement

Hunt Lead Grid

Hunt Lead Level Hunt Lead Name ATT&CK Risk of False Positives
1 (Junior) Campaign IP Sweep for ARTEX Servers and WISP C2 T1071.001 (Application Layer Protocol: Web Protocols) Low. The addresses are attacker-operated servers Microsoft observed through Nov. 6; the main risk is reassignment of rented VPS addresses later, so anchor on dates from mid-October 2026 onward.
2 (Pro) curl or netcat Pulling a First-Seen File onto a Server T1105 (Ingress Tool Transfer) Medium. Admins, package scripts and CI jobs use curl constantly; the lead anchors on a remote file the host has never fetched before, on internet-facing servers, and on curl or nc whose parent is a web or application service.
2 (Pro) Web Application Process Spawning a Shell After Upload or SSRF Abuse T1190 (Exploit Public-Facing Application), T1505.003 (Server Software Component: Web Shell) Medium. Some web apps legitimately shell out to image tools or converters; the lead anchors on interactive shells, download utilities or recon commands as children of the web server process, and on new script files in upload directories.
3 (Elite) Machine-Speed Scan-and-Retry Bursts from Default Tool Clients T1595.002 (Active Scanning: Vulnerability Scanning) High on its own. Internet background scanning never stops; the lead only becomes actionable when a burst is followed by a successful exploit-style request, a decoy hit or the download behavior in the earlier leads.

Campaign IP Sweep for ARTEX Servers and WISP C2

Microsoft lists three WISP C2 nodes, 199.19.95[.]66, 199.19.95[.]67 and 199.19.95[.]68 (first seen Nov. 2, 2026), four ARTEX servers, 91.92.240[.]238 and 85.11.168[.]45 (first seen Nov. 4), 178.16.54[.]129 (first seen Oct. 19) and 31.77.229[.]35 (first seen Oct. 28), and a C2 server, 85.11.167[.]52 (first seen Nov. 4), all active as of Nov. 6. Inbound connections from the ARTEX servers point to scanning and exploitation of your internet-facing systems; outbound connections to the WISP nodes or the C2 server mean a RAT is already running. The detection logic is a retrospective sweep of every network log you hold for these addresses in both directions, using Microsoft’s published Sentinel query below. Note that the published query spells one ARTEX server 178.15.54.129 while the indicator table gives 178.16.54[.]129; add both to your own copy until Microsoft clarifies.

Use firewall, NetFlow, proxy, load balancer and web server access logs, plus EDR network events on servers; the ASIM network session parser in the query covers most of these in Sentinel. Sweep from Oct. 1, 2026, since the earliest server was first seen Oct. 19 and Microsoft says the activity began in October. Scope inbound hits to internet-facing web applications and outbound hits to every server. Microsoft cautions that the group uses proxies and VPNs, so a clean result doesn’t clear you; pair this sweep with the behavioral leads. Block the exact IPs rather than the hosting ranges they sit in. Any outbound hit to a WISP node or the C2 server is a compromised host and goes straight to incident response; inbound-only hits should be checked against the web shell lead for that server.

let lookback = 30d; let
ioc_ip_addr = dynamic(["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", "85.11.167.52"]);
_Im_NetworkSession(starttime=todatetime(ago(lookback)), endtime=now())
| where DstIpAddr in (ioc_ip_addr)
| summarize FirstSeen=min(TimeGenerated), LastSeen=max(TimeGenerated), EventCount=count() by SrcIpAddr, DstIpAddr, DstDomain, Dvc, EventProduct, EventVendor

(Microsoft Sentinel query reproduced as published by Microsoft Threat Intelligence.)

Hunt Lead Level: 1 (Junior Analyst)

ATT&CK: Command and Control → T1071.001 (Application Layer Protocol: Web Protocols) (our mapping; the source does not state the RAT protocols)

Sources: “ARTEX abuse expands as cybercriminals target financial sector”, Microsoft Threat Intelligence, Nov. 6, 2026, https://www.microsoft.com/en-us/security/blog/2026/11/06/artex-abuse-expands-as-cybercriminals-target-financial-sector/ (source article)

curl or netcat Pulling a First-Seen File onto a Server

After exploitation, Microsoft observed ARTEX agents using netcat and curl to download the WISP, vshell and AdaptixC2 RATs onto compromised systems. This is the step where an agent’s speed stops helping it: a RAT has to arrive from somewhere, and on a server that somewhere is usually a URL or bare IP the host has never contacted. Microsoft’s published Defender XDR query builds a seven-day baseline of every remote file or IP that curl fetched on each host (from both Defender network events and Windows 4688 process creation), then returns curl downloads in the last day that fall outside that baseline. Build on it by adding nc, ncat and wget, by flagging downloads to world-writable paths such as /tmp, /dev/shm or C:\Users\Public followed by a chmod +x or an execution of the downloaded file, and by weighting hits where the parent process is a web server, application runtime or service account.

Run the query in Defender XDR advanced hunting or Sentinel, and add Linux process telemetry (auditd execve or EDR) for internet-facing servers, since Storm-3358 targets web applications that often run on Linux. Sweep as far back as your process data allows, at least to Oct. 1, 2026. Scope first to internet-facing servers and anything in a DMZ, where interactive curl use by humans is rare. Expect hits from package managers, update scripts and CI runners; allowlist by full command line and destination for those jobs rather than by process name, because the attacker uses the same binaries. Blocking curl or wget outright is the wrong control on Linux servers and breaks maintenance. A first-seen download by a web service account, followed by execution, should be escalated as a likely RAT install and correlated with the IP sweep.

let known_files = DeviceNetworkEvents
    | where TimeGenerated between (ago(7d)..ago(1d))
    | where InitiatingProcessFileName has "curl"
    | extend url = extract("http[s]?:\\/\\/(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\\(\\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+", 0,InitiatingProcessCommandLine)
    | extend ip = extract("(\\b25[0-5]|\\b2[0-4][0-9]|\\b[01]?[0-9][0-9]?)(\\.(25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)){3}[^ ]*", 0, InitiatingProcessCommandLine)
    | extend remote_file = iif(isnotempty(url), url, ip)
    | union (SecurityEvent
    | where TimeGenerated between (ago(7d)..ago(1d))
    | where EventID == 4688
    | where CommandLine has "curl"
    | extend url = extract("http[s]?:\\/\\/(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\\(\\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+", 0,CommandLine)
    | extend ip = extract("(\\b25[0-5]|\\b2[0-4][0-9]|\\b[01]?[0-9][0-9]?)(\\.(25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)){3}[^ ]*", 0, CommandLine)
    | extend remote_file = iif(isnotempty(url), url, ip))
    | summarize by remote_file;
    DeviceNetworkEvents
    | where TimeGenerated > ago(1d)
    | where InitiatingProcessFileName has "curl"
    | extend url = extract("http[s]?:\\/\\/(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\\(\\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+", 0,InitiatingProcessCommandLine)
    | extend ip = extract("(\\b25[0-5]|\\b2[0-4][0-9]|\\b[01]?[0-9][0-9]?)(\\.(25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)){3}[^ ]*", 0, InitiatingProcessCommandLine)
    | extend remote_file = iif(isnotempty(url), url, ip)
    | union (SecurityEvent
    | where EventID == 4688
    | where CommandLine has "curl"
    | extend url = extract("http[s]?:\\/\\/(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\\(\\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+", 0,CommandLine)
    | extend ip = extract("(\\b25[0-5]|\\b2[0-4][0-9]|\\b[01]?[0-9][0-9]?)(\\.(25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)){3}[^ ]*", 0, CommandLine)
    | extend remote_file = iif(isnotempty(url), url, ip))
    | where remote_file !in (known_files)

(Microsoft Defender XDR advanced hunting query “Download of new files using curl”, reproduced as published by Microsoft Threat Intelligence.)

Hunt Lead Level: 2 (Pro)

ATT&CK: Command and Control → T1105 (Ingress Tool Transfer) (our mapping)

Sources: “ARTEX abuse expands as cybercriminals target financial sector”, Microsoft Threat Intelligence, Nov. 6, 2026, https://www.microsoft.com/en-us/security/blog/2026/11/06/artex-abuse-expands-as-cybercriminals-target-financial-sector/ (source article)

Web Application Process Spawning a Shell After Upload or SSRF Abuse

Storm-3358’s way in is web application flaws, specifically anonymous file upload and server-side request forgery, and Microsoft says ARTEX intrusions use web shells for persistence on compromised web servers. Microsoft’s Defender coverage for this activity includes “Suspicious shell execution from PHP application”, “Possible Web Server compromise activity”, “Suspicious process executed by network service” and “Suspicious cron job creation”, which tells you where the behavior lands on the host. The detection logic has two halves: a new script file (PHP, JSP, ASPX or similar) written into an upload or web-served directory by the web server process, and the web server or application runtime (php-fpm, httpd, nginx workers, Tomcat, w3wp) spawning sh, bash, cmd, PowerShell, curl, wget, nc or recon commands such as id, whoami and uname. For SSRF, add outbound requests from the application tier to internal-only addresses or cloud metadata endpoints that the app never normally calls.

Use EDR process lineage on web servers, file integrity monitoring on upload and webroot directories, web server access logs (POSTs to upload handlers followed by GETs to the uploaded file), and application-tier egress logs for the SSRF half. Sweep from Oct. 1, 2026 and keep the rule running. Scope to internet-facing web and API servers, especially those of payment and banking front ends. Some applications legitimately call ImageMagick, ffmpeg or PDF converters; allowlist those by full child command line, never by parent. A WAF rule for one upload path is the wrong end state, because the agent will try the next path within seconds. A hit is a compromised server: preserve it, remove the web shell, rotate every credential the app held, and pivot to the curl download lead on the same host.

Hunt Lead Level: 2 (Pro)

ATT&CK: Initial Access → T1190 (Exploit Public-Facing Application); Persistence → T1505.003 (Server Software Component: Web Shell) (our mapping)

Sources: “ARTEX abuse expands as cybercriminals target financial sector”, Microsoft Threat Intelligence, Nov. 6, 2026, https://www.microsoft.com/en-us/security/blog/2026/11/06/artex-abuse-expands-as-cybercriminals-target-financial-sector/ (source article)

Machine-Speed Scan-and-Retry Bursts from Default Tool Clients

Microsoft gives the behavioral signature of an agent on the doorstep: a wide variety of requests in a brief time, retries of variations on the same technique, and moves to next-stage actions at machine speed. By default ARTEX drives Nmap, curl, wget, Python HTTP clients, a WebFetch SDK and headless Chromium through Playwright, so early traffic often carries those tools’ default user agents. The detection logic is stateful: per source IP, count distinct URL paths, parameters and payload variants against a single application within a short window (for example 50 or more distinct exploit-style requests in five minutes), flag clusters where the same endpoint is retried with small mutations, and raise severity when the burst is followed within minutes by an HTTP 200 on an upload or URL-fetch endpoint, a request to a decoy path, or the download and web shell behavior in the earlier leads. Default tool user agents are a supporting feature, not the trigger.

Use WAF and reverse proxy logs with full request lines and user agents, web server access logs and, if you run them, honeypot or decoy telemetry; Microsoft reports that honeypots, tarpits and fake-win decoys reduced attack traffic reaching real assets and produced high-confidence signal in its testing. Baseline normal crawler and monitoring traffic for at least 30 days and maintain an allowlist of your own scanners, uptime checks and partners by IP. Expect constant background scanning; that is why this lead stays at Elite level and why it must be chained to a follow-on event before anyone is paged. Blocking default user agents is the wrong primary control, as Microsoft notes the agent will simply change them. A burst that ends in a successful exploit-style request should trigger the web shell and download leads on that server immediately. The pattern generalizes to any LLM-driven attack tool working through targets by trial and error.

Hunt Lead Level: 3 (Elite)

ATT&CK: Reconnaissance → T1595.002 (Active Scanning: Vulnerability Scanning) (our mapping)

Sources: “ARTEX abuse expands as cybercriminals target financial sector”, Microsoft Threat Intelligence, Nov. 6, 2026, https://www.microsoft.com/en-us/security/blog/2026/11/06/artex-abuse-expands-as-cybercriminals-target-financial-sector/ (source article)

Hash Appendix

The source article publishes no file hashes for the WISP, vshell or AdaptixC2 payloads; its indicators are IP addresses only. No hashes were added from other sources.

References Appendix

  • “ARTEX abuse expands as cybercriminals target financial sector”, Microsoft Threat Intelligence, Nov. 6, 2026, https://www.microsoft.com/en-us/security/blog/2026/11/06/artex-abuse-expands-as-cybercriminals-target-financial-sector/ (source article; activity description, indicators, Defender detections and hunting queries)
  • MITRE ATT&CK Enterprise technique pages linked above, MITRE, https://attack.mitre.org/

More from the Roundup