AI Agents Speed Up Supply Chain Breaches

Microsoft confirmed this week that a China-aligned hacking group utilized an open-source AI framework to execute a large-scale supply chain attack. By exploiting stolen credentials from a third-party application, the attackers managed to attempt sign-ins across more than 2,250 Microsoft 365 tenants. The speed of this operation was notable, with some intrusions reaching privileged access levels in under three hours from the initial targeting phase.

The core of this operation involved a tool called ARTEX, a system designed to run autonomous penetration tests using large language models. Instead of relying solely on human intuition, the attackers deployed AI agents to handle scanning, reconnaissance, and password guessing. Microsoft notes that these agents can operate around the clock with minimal human intervention, creating a pace that leaves defenders with very little time to detect anomalies and respond effectively. This automated approach significantly accelerated the abuse of OAuth applications and the scanning of vulnerable systems.

In addition to using stolen application credentials, the group engaged in automated password spraying. This technique targeted 327 accounts across eight different organizations, resulting in over 151,000 failed login attempts. Despite the high volume of attempts, the attackers did not report any successful logins through this specific method. The sheer scale of the automation suggests a shift toward using AI to brute-force access at a speed and volume that traditional manual attacks rarely achieve.

The hacking group, identified as Storm-0940, has been active since 2021 and is described as a developing actor with no known industry aliases. While the group targets various sectors, this specific campaign focused on aerospace and defense, semiconductors, government agencies, and think tanks. Microsoft clarified that the breach did not stem from a vulnerability in its own systems or products, but rather from unauthorized access to a multi-tenant application used by the victims.

At the peak of the password spraying campaign on October 11, the system generated 2,900 failed attempts per minute, totaling more than 111,000 for that day alone. Nearly 98% of these activities resulted in account lockouts. However, the attackers also used four specific application credentials to sign in to customer tenants from the same IP range. In one instance, they achieved over 200 sign-ins in under four minutes, a speed indicating a script working systematically through a target list.

Security firm Huntress observed specific behaviors in one tenant on October 15, where an application added itself to the top administrator role. This action triggered alerts regarding self-assigned global administrator privileges. The attackers also planted new secrets with expiration dates set for 2028. By October 16, Huntress had confirmed unauthorized sign-ins for 162 tenants, and by the next day, nearly 1,000 tenants showed sign-in activity from the attacking IP addresses, with more than 300 of those attempts succeeding.

The use of ARTEX marks a distinct departure from traditional attack methods, as the framework is open-source and compatible with various AI models. Public reports this month have linked similar tools to breaches at South Korean financial institutions, with one suspect allegedly using a specific AI coding tool. The developer of ARTEX has announced that the project will no longer be updated and will transition to closed-source status, potentially impacting future threat actor strategies.

ConnectWise, the affected platform, suspended several of its services, including Cloud Backup and SaaS Security, as a precautionary measure. A phased restoration process has begun, 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 instructions are issued. Organizations are encouraged to review their activity logs from this month to identify any unauthorized access.