AI Agents Drive Rapid Microsoft Breaches

The most significant development in this unfolding security incident is the confirmation that artificial intelligence agents were deployed to accelerate a large-scale breach. According to Microsoft, the Storm-0940 actor group utilized ARTEX, an open-source agentic penetration testing framework, to automate reconnaissance and exploitation. This automation allowed the group to attempt sign-ins across more than 2,250 Microsoft 365 tenants, with some attacks achieving privileged access in less than three hours.

The incident began after the group gained unauthorized access to ConnectWise, a multi-tenant application. By leveraging stolen credentials for ConnectWise apps, the attackers could authenticate without interactive user sign-ins. This method bypassed traditional user-based defenses. The speed of the operation was notable; the AI agents operated continuously with minimal human interaction, a pace that left defenders with little time to detect or respond to the intrusion.

Prior to the main breach, the group executed an automated password spraying campaign. This activity targeted 327 accounts across eight organizations, resulting in over 151,000 failed logon attempts. At its peak on October 11, the spray campaign generated 2,900 failed attempts per minute. Although nearly 98% of these attempts were locked out, the subsequent use of compromised application credentials allowed the attackers to move laterally.

The targeted industries included aerospace, defense, semiconductor, government, and think tanks. Microsoft clarified that the breach did not stem from a vulnerability in its own systems, but rather from the compromise of the ConnectWise platform. Storm-0940, active since 2021, is identified as a China-state-aligned actor. The group’s use of a script to sweep through tenants in sorted order, with sign-ins occurring seconds apart, indicates a highly organized, automated approach.

Security firm Huntress observed specific behaviors within compromised tenants, such as applications adding themselves to top administrator roles. This triggered alerts for self-assigned Global Administrator privileges. By October 16, Huntress had confirmed unauthorized sign-ins for 162 tenants. The following day, the number of Entra tenants showing sign-in activity from the attacking IP reached nearly 1,000, with more than 300 successful logins recorded.

The use of ARTEX distinguishes this attack from many traditional methods. The framework, which can utilize any AI model, has also been linked to recent breaches at South Korean financial institutions. In those separate incidents, CrowdStrike attributed the use of ARTEX, specifically with Anthropic’s Claude Code, to a likely China-based suspect. The developer of ARTEX has since announced on GitHub that the project will no longer be updated and will transition to closed source.

In response, ConnectWise suspended its Cloud Backup, SaaS Security, and Cloud Management services as a precaution. The company has begun a phased restoration process, with recovery expected to take several days. While there are no current indicators that partner backup data was compromised, Microsoft advises customers to keep affected app connections disabled until formal restoration instructions are issued. Organizations are also urged to review activity logs from this month to identify potential exposure.

The incident highlights the growing role of autonomous AI in cyberattacks. The ability of these agents to work around the clock and rapidly escalate privileges presents a new challenge for security teams. As the restoration process continues, the full extent of the compromise remains uncertain, particularly regarding how many tenants were actually breached and what specific data was accessed.