Microsoft Threat Intelligence has tied the agentic penetration-testing framework Artex to a criminal operator tracked as Storm-3358, and the target set has shifted from espionage toward finance.
The group combines Artex with ordinary automation to sweep internet-facing systems, exploit anonymous file upload and server-side request forgery bugs, and drop open-source remote access trojans. Microsoft named three in use: WISP, vshell and AdaptixC2, the last serving for persistence and lateral movement.
What the framework does once inside
Artex agents scan and exploit vulnerable web applications, leave web shells behind for persistence, harvest credentials from public sources and already-compromised machines, and pivot through stolen cloud identities to pull data out. Brute-force password guessing is on the list too.
The tell for defenders is speed and variety. An agentic framework throws a wide spread of requests in a very short window, retries variants of the same technique, and jumps to the next stage at machine pace. It also reuses familiar pentest tooling, which makes signature work harder.
The GitHub reconnaissance layer
Separate reporting describes agents querying public repositories for exposed secrets, then using still-valid application credentials against downstream Microsoft tenants. One address began signing in that way from October 28, reaching Key Vaults, Storage Accounts and Azure AI services in at least one case.
Microsoft’s guidance is the uncomfortable part: if a secret appeared in a public repo, treat it as compromised. Detection advice runs toward deception rather than pure blocking. Honeypots, decoy surfaces and tarpits give a machine-speed attacker something to spend its budget on.
