Storm-0940 Used AI Agents to Hammer ConnectWise Customers at 2,900 Login Attempts Per Minute
Microsoft has named Storm-0940, a China state-aligned threat actor, as responsible for the recent supply chain attack against ConnectWise's cloud backup service. The detail that makes this one worth paying attention to: the attackers used an AI-powered penetration testing framework called ARTEX to automate the dirty work at a pace no human operator could match.
ARTEX is a Go-based agentic framework originally built for offensive security research. It was quietly made closed-source after it turned up in attacks against South Korean financial institutions. Apparently that didn't slow anyone down much.
Microsoft says Storm-0940 has been running ARTEX-assisted campaigns since at least September 2026, targeting organisations across aerospace, defence, energy, financial services, government, manufacturing, semiconductors, and a dozen other sectors. The AI tooling handled scanning, reconnaissance, password spraying, and OAuth application abuse. Automatically. Continuously. Without needing a human to babysit the process overnight.
The ConnectWise incident itself is a useful case study in how quickly things can go wrong. Between October 10 and 14, Storm-0940 infrastructure hammered 327 accounts across eight organisations, generating over 151,000 failed login attempts. The peak was October 11, when the system was generating 2,900 failures per minute. That's not a slow, stealthy attack. That's a battering ram. Microsoft detected no successful authentications during this window.
Then October 15 arrived. The AI agents got hold of credentials for multiple ConnectWise cloud application identities and used them to authenticate directly into customer environments. Because these applications already had established trust relationships within customer tenants, there was no need for interactive sign-ins. The agents just walked in through the front door using someone else's key.
From there, in some cases, the compromise extended to Microsoft Graph access, directory operations, credential modifications, and privilege escalation attempts. Microsoft notes that changes made to application credentials can persist as backdoors even after the original compromised app is cleaned up. So remediating the obvious entry point doesn't necessarily mean the threat is gone.
The speed is what stands out. Microsoft observed cases where the gap between initial targeting and achieving privileged cloud access was under three hours.
Storm-0940 isn't the only crew using ARTEX. Microsoft says it has spotted the framework being used by threat actors across multiple regions. Recorded Future's Insikt Group is tracking a related cluster under the name TAG-217, with activity traced back to mid-August 2026. That group was also experimenting with attacks on cryptocurrency mining infrastructure and probing weaknesses in mobile operating systems.
Zscaler ThreatLabz dug into a specific ARTEX deployment and found some telling configuration details. The setup used a DeepSeek model through an OpenAI-compatible API gateway, with other configurations referencing GLM and Grok. The multi-model setup was deliberate: prompts explicitly instructed one AI to review the outputs of another. GLM reviewing DeepSeek's findings. AI peer review for cyberattacks.
Prompt history analysis suggested financial motivation. Requests included bulk extraction of personal and transaction data, questions about offloading stolen data, payment gateway API key extraction, and exploiting NFT minting workflows. Not exactly subtle.
ConnectWise temporarily shut down services following the incident and has been restoring them in phases. The company acknowledged that nation-state actors are increasingly treating the software supply chain as a primary target precisely because a single trusted connection can be a skeleton key to dozens of downstream organisations.
Microsoft put it plainly: AI agents don't take weekends off, don't need sleep, and can run multiple attack paths simultaneously. The gap between picking a target and achieving objectives is shrinking. Supply chain attacks just got faster and cheaper to execute at scale.