AI Agents Exploit Microsoft 365 via ConnectWise Credentials

Hackers have utilized AI agents and compromised ConnectWise credentials in attempts to infiltrate over 2,250 Microsoft 365 tenants, as outlined in a recent technical report by Microsoft. The cyber campaign is attributed to Storm-0940, a group believed to be aligned with the Chinese state. This attack highlights the vulnerabilities in current cybersecurity frameworks, though not every tenant was necessarily breached.

Exploit of ConnectWise Permissions

The cyberattack capitalized on permissions that were already granted to trusted ConnectWise applications. Earlier reports indicated unauthorized sign-ins across 173 partner and customer tenants due to compromised backup credentials. The attackers bypassed typical user authentication barriers by authenticating as applications, thus avoiding standard multi-factor authentication prompts.

The credential abuse was initially identified by researchers at Huntress, who conducted a technical analysis on October 26. Their findings corroborate Microsoft’s assessment of the situation but do not independently confirm AI involvement. This report specifically addresses malicious application activities and attack tools, rather than identifying a new malware variant.

Details on ARTEX Framework and Attack Methods

The ARTEX framework, employed in this attack, coordinates multiple AI agents to automate tasks ranging from reconnaissance to data exfiltration. Microsoft acknowledges that this framework can create a comprehensive map of discovered vulnerabilities and assets. The use of stolen application credentials allowed hackers to exploit existing customer trust, rather than exploiting a newly discovered Microsoft vulnerability.

Huntress noted significant activity on October 15, with one automated sweep processing nearly 1,000 tenants within a short timeframe. The attempts included bursts of failed token requests, indicating parallel operations. Changes in attack strategy were also observed, such as retry logic after failed certificate uploads, showcasing the adaptability of their tools.

Implications for Microsoft 365 Security

The campaign has exposed gaps in Microsoft 365’s security, particularly due to missing Microsoft Graph activity logs. Without these logs, investigators face challenges in determining which information was accessed or stolen. Microsoft differentiates between targeting, successful logins, and confirmed data access, with earlier investigations recording over 151,000 failed logins across eight organizations.

The incident underscores the growing threat posed by AI agents in cybersecurity. While the focus remains on harmful actions rather than AI assumptions, Huntress advises monitoring for unexpected application behaviors and cross-tenant campaigns. Microsoft recommends disabling affected connections, preserving evidence, and reviewing credentials and permissions thoroughly.

Organizations are advised to enhance logging capabilities and remain vigilant for suspicious activities. Continuous monitoring and proactive security measures are essential to mitigate such threats going forward.