AI Pentesting Tool ARTEX Linked to China-Backed ConnectWise Breach, Microsoft Says

What Microsoft Found

Microsoft’s report names Storm-0940 as the actor behind the ConnectWise supply chain compromise. The group used stolen ConnectWise credentials as an entry point, then deployed ARTEX — an open-source, AI-driven pentesting framework written in Go — to automate and scale the attack across hundreds of organizations.

The sectors targeted include aerospace, defense, energy, financial services, government, and manufacturing. The numbers Microsoft published give a clear picture of the scale:

• 2,250+ Microsoft 365 tenants reached
• 2,900 sign-in attempts per minute at peak
• 151,000+ failed sign-ins recorded
• 327 accounts with app-only sign-ins affected by password spray
• 200+ sign-ins completed in under four minutes

These are not the metrics of a manual operation. They reflect automated agents running continuously without pause.

How the Attack Unfolded

Storm-0940 combined several techniques in sequence. The initial phase relied on password spray attacks targeting app-only sign-ins — accounts that authenticate without interactive user sessions, making them harder to catch through conventional monitoring.

Once authentication succeeded in a subset of environments, the group moved into Microsoft Graph to run directory reconnaissance. From there, they modified application credentials to establish persistence and escalated to privileged roles across multiple customer environments.

The persistence mechanism is the detail that demands immediate attention. According to Microsoft, credential modifications made to trusted applications can remain active as a backdoor even after the original compromised application has been remediated. An organization that disabled its ConnectWise connection after the public breach announcement may still have an active entry point if application credentials were modified beforehand.

OAuth Abuse and the Persistence Problem

The OAuth abuse component allowed Storm-0940 to operate without requiring interactive user sign-ins. By authenticating through applications that already held granted permissions, the attackers entered customer environments in a way that bypasses many standard detection triggers.

Microsoft’s framing is direct: remediation of the ConnectWise connection is a starting point, not a resolution. Any application credential changes made during the window of access need to be audited and revoked independently.

ARTEX Is Not Exclusive to Storm-0940

Microsoft notes that ARTEX has been used by threat actors across multiple regions, not only by Storm-0940. The framework is open-source and can be configured to use different AI models for planning and executing tasks. Researchers at Zscaler and Recorded Future have separately identified a Chinese-speaking group that coordinated use of the tool with models including DeepSeek, GLM, and Grok.

The public availability of ARTEX means that removing it from one group’s hands does not remove it from the threat landscape. Any actor with a copy of the framework can replicate this attack pattern.

This adds to broader concerns around AI-driven pentesting frameworks and how offensive workflows can be scaled.

Why Speed Is the Central Problem

Microsoft returns repeatedly to the speed dimension in its report. Agentic AI does not sleep, does not take weekends, and can run multiple attack threads simultaneously. At 2,900 sign-in attempts per minute, manual detection and response cannot keep pace.

This activity demonstrates how threat actors are integrating agentic AI into existing offensive security workflows and allows them to scale their operations exponentially.

This is not a theoretical risk. It is a documented, active attack pattern with measurable outcomes across more than 2,250 organizations.

The core implication is the need for machine-speed detection as attack automation accelerates.

What Security and IT Teams Should Do Now

The persistence mechanism is the most urgent item. Even if your organization has already addressed the ConnectWise exposure, the following steps remain relevant:

• Audit application credentials modified during the breach window, not just the ConnectWise application itself
• Review OAuth permissions granted to third-party applications across your Microsoft 365 environment
• Check for privileged role assignments that were not explicitly authorized
• Examine Microsoft Graph activity logs for directory reconnaissance patterns
• Treat October 15 as a disclosure date, not a remediation date — the actual window of access may have started earlier

The broader takeaway is structural. Machine-speed attacks require machine-speed detection. Organizations still relying on manual review cycles for cloud authentication anomalies are operating with a response gap that adversaries like Storm-0940 are actively exploiting.