Skip to content
nulltapE-reader edition

← All e-reader articles

4 min read

Microsoft Tracks ARTEX Web Intrusions Against Financial Firms

Storm-3358 exploited exposed upload and SSRF weaknesses, then installed remote access tools. Trace web-server activity into endpoint and identity logs.

Read on the standard site

Illustration of an exposed server access hatch in a financial operations room, with red network pulses entering it.

Storm-3358 began targeting financial organizations with the ARTEX agentic offensive framework in October, according to Microsoft Threat Intelligence. Microsoft observed the group scanning internet-facing systems, exploiting anonymous file upload and server-side request forgery (SSRF) weaknesses, and installing remote access tools. The reported targets are mainly financial organizations in Southeast Asia, with a smaller number of large financial organizations elsewhere.

A web application exposed to the internet is the entry point to prioritize. Once an attacker gets execution on its server, a legitimate-looking utility download can be the step that brings in persistent access. Microsoft identified WISP, vshell and AdaptixC2 in this activity. That sequence gives defenders a more useful starting point than looking for a single “AI attack” signature.

Web flaws lead to remote access

Microsoft says Storm-3358 combined ARTEX with conventional automated tools. Its agents scanned public-facing applications and exploited flaws including anonymous upload and SSRF. On compromised systems, Microsoft observed the agents using netcat and curl to download the named remote access tools. In recent ARTEX-related attacks across groups, Microsoft also observed web shells and credential theft; compromised cloud identities can support persistence and data exfiltration. Check those paths on a Storm-3358 host when its own activity supports the search.

The uploaded-file and SSRF paths deserve separate attention. An upload path that accepts anonymous files gives the attacker a place to put unwanted content; an SSRF path can make a server request a destination on the attacker’s behalf. The specific vulnerable applications and configurations are undisclosed, so inventory your exposed upload endpoints and server-side fetch features before assuming a particular product needs a patch. Limit anonymous upload and constrain outbound destinations on services where those capabilities are unnecessary.

Google Threat Intelligence Group and Mandiant have separately observed ARTEX in incident response engagements across several threat groups. Their research describes compressed activity bursts and iterative trials, useful context for hunting. Use Google’s burst pattern to hunt ARTEX activity; scope Storm-3358 against Microsoft’s financial-sector observations.

Fast requests leave joined-up traces

Microsoft describes ARTEX scanning as a “wide variety of requests in a brief time,” often followed by quick variations on one technique or an immediate next action. Start with web access logs for the exposed applications, then line up response codes, upload attempts and outbound requests against process starts on the same host. Look for curl or netcat launched by a web-service process and for the Defender for Endpoint alerts Suspicious shell execution from PHP application, Possible Web Server compromise activity, Suspicious curl behavior and Possible reverse shell. These detections are leads: a legitimate deployment can also use curl.

Microsoft published sample network indicators. Its table lists WISP command-and-control nodes 199.19.95[.]66, 199.19.95[.]67 and 199.19.95[.]68; ARTEX servers 91.92.240[.]238, 85.11.168[.]45, 178.16.54[.]129 and 31.77.229[.]35; and a command-and-control server 85.11.167[.]52. They were listed as active on November 6. Compare historical network sessions and affected web hosts with those values, then follow the associated process and sign-in history. Microsoft warns that proxies and VPNs change attacker source IPs. Confirm an address match against activity on the affected web host.

The example Sentinel query in Microsoft’s post uses 178.15.54[.]129 where its indicator table lists 178.16.54[.]129. Treat the table value as the reported indicator and reconcile that difference before importing the query as a block rule. The short-lived source IP is less durable than the sequence from web request to host execution and new access.

Follow access into cloud identities

Microsoft’s separate November 5 account describes ARTEX-assisted searches for exposed application credentials and follow-on service-principal sign-ins. That account covers a different activity path. For a web host in this investigation, expand the search into cloud identities when its own credential exposure or subsequent authentications warrant it.

Microsoft’s Entra sign-in documentation identifies separate service-principal and non-interactive user sign-in views in addition to interactive sign-ins. If the compromised web service holds application credentials, correlate the service account, application ID, resource, source IP and sign-in time with the server’s process and outbound-network events. An unexpected successful token request following web-shell or RAT activity calls for containment of that credential and a review of what it accessed.

Contain the web-to-identity path

For financial-sector web services exposed to the internet, first find anonymously writable upload endpoints and server-side fetch features. Restrict unnecessary anonymous upload, constrain outbound destinations and fix confirmed application weaknesses. On a host with matching activity, isolate the affected system, remove unauthorized web shells or remote access tools through incident response, and rotate credentials it held. Search the server’s web, endpoint and outbound-network logs alongside Entra service-principal sign-ins where that server used cloud credentials.

Check the resulting state at the same boundaries the attacker crossed: the exposed upload path rejects unauthenticated writes where they are not required; the fetch feature reaches only approved destinations; the affected host no longer starts unapproved remote access processes; and new sign-ins by its rotated identity originate from the intended workload.

Primary sources