Storm-0940 Used AI Framework to Target Hundreds of Organizations; ConnectWise Supply Chain Breached
Storm-0940 compromised ConnectWise's Cloud Backup service by obtaining an API secret that granted Microsoft Graph privileges to read and modify credentials across customer tenants. The group then used ARTEX, an open-source agentic AI framework, to automate reconnaissance and credential spraying at scale: 151,000 failed logon attempts across eight organizations over four days (peaking at 2,900 per minute), followed by pivoting to ConnectWise customers on October 15. CyberScoop and Huntress data show the actual compromise scope remains contested—Huntress observed sign-in activity across "almost 1,000 Entra tenants" with "more than 300" successful sign-ins by late Saturday, while confirming unauthorized access in 162 of its own customer tenants; Microsoft's attribution cites "more than 2,250 Microsoft 365 tenants" reached by the ARTEX campaign. The discrepancy likely reflects Huntress's visibility into a subset of customers versus Microsoft's telemetry across the full attack surface, but the upper-bound figure (2,250) should be treated as the operational scope until Microsoft publishes a forensic breakdown.
The actual damage vector runs through OAuth token abuse and credential modification—not data exfiltration per se. ConnectWise's API secret granted the attacker standing to modify application credentials and establish persistence in victim environments without triggering perimeter defenses, because the requests originated from a trusted third-party connector. This means defenders cannot contain a ConnectWise compromise at the firewall or email gateway; they must assume every downstream customer's identity directory, application permissions, and token store require forensic review and rotation. For organizations in aerospace, defense, semiconductors, and financial services that rely on MSP-managed backup and identity connectors, this creates a cascading liability: a single vendor's credential leak now mandates organization-wide credential rotation, forensic review of OAuth token usage, and temporary suspension of automated backup and management services.
Storm-0940 demonstrated that an open-source AI penetration-testing framework can compress a nation-state's attack tempo from hours per target to minutes per thousand targets, and that a single compromised third-party connector with Microsoft Graph privileges can tunnel into the identity and credential systems of thousands of downstream customers simultaneously.
The actual intrusion count remains uncertain—Huntress confirms 162 unauthorized accesses in its own customer base, while Microsoft's telemetry suggests 2,250 tenants were reached by ARTEX, but the number of tenants where data was actually exfiltrated (versus authentication-only success) has not been disclosed. The second-order effect is not the immediate data theft but the credential rotation and forensic review burden now cascading to every customer of ConnectWise and, by extension, every organization using similar MSP connectors for backup, identity management, or cloud security.
This inverts the current assumption that third-party risk can be managed through vendor contracts and periodic audits; it now requires that every organization maintain forensic telemetry on third-party connector activity and be prepared to rotate credentials at scale within hours of a breach notification. Watch whether Microsoft, Okta, or other identity-platform vendors announce new conditional-access or token-filtering capabilities designed to isolate third-party connectors—the absence of such announcements would indicate that the industry is not yet moving to address the architectural vulnerability exposed by this attack.
How many of the 2,250 compromised tenants actually had data exfiltrated, versus just authentication success? Microsoft said successful authentication doesn't necessarily indicate data theft — what's the actual intrusion count?
Fault Lines
Strategic intelligence, synthesized daily — with a public track record. Every call graded against what actually happened.