AI Automation in Cyber Intrusion: Microsoft Blames Chinese Group for Massive Breach

Microsoft has officially linked a widespread supply chain intrusion to Storm-0940, a threat actor it associates with the Chinese state. The breach, which targeted a major business software provider, leveraged an open-source artificial intelligence tool designed for automated security testing. This attribution highlights a growing trend where attackers utilize large language model agents to accelerate the pace of cyber operations beyond the speed of traditional human-led campaigns.

The core of the incident involved unauthorized access to a multi-tenant application environment. From that foothold, the group attempted logins across more than 2,250 different organizations using Microsoft 365. In several cases, the intruders secured privileged access within three hours of their initial targeting. The attackers exploited stolen credentials that allowed them to authenticate as the software applications themselves, bypassing the need for interactive user sign-ins. While the method of obtaining these credentials remains unknown, the efficiency of the takeover was stark.

The tool used, known as ARTEX, functions as a system of agents that run autonomous penetration tests. Microsoft notes that this framework sped up reconnaissance, password spraying, and the abuse of authorization protocols. Because these agents can operate continuously with minimal human oversight, defenders face a compressed window for detection and response. The targeted sectors were broad, spanning fourteen industries including aerospace, semiconductors, and government bodies, though Microsoft clarified that its own core products were not the source of the vulnerability.

A separate but parallel effort involved aggressive password spraying across eight organizations. This campaign generated over 151,000 failed login attempts without achieving a single success. At its peak on October 11, the rate hit 2,900 attempts per minute. Despite the volume, nearly 98 percent of the activity resulted in account lockouts, indicating that this specific method was largely ineffective compared to the credential-based approach.

The more successful vector involved the identity takeover of several cloud applications. Using just four sets of credentials, the AI agents recorded over 200 sign-ins in under four minutes from a single IP range. They performed directory operations and modified privileged roles. Security firm Huntress observed one tenant on October 15 where an application added itself to the highest administrator role. By October 17, Huntress had identified sign-in activity from the attacking IP at nearly 1,000 tenants, with over 300 of those sign-ins successful. The rapid, sequential movement through the list suggested a script-driven operation.

The use of an AI-driven framework that works with various models sets this campaign apart from typical attacks. Public reporting has linked similar tools to breaches in South Korea. Meanwhile, the developer of ARTEX announced on GitHub that the project would cease updates and become closed source. The software provider has suspended several services as a precaution and begun a phased restoration, stating it has no current indicators that partner backup data was taken. Customers are advised to keep affected connections disabled until further notice.

The incident underscores the shifting landscape of enterprise security. As automated agents become more prevalent, the speed of attack increases, demanding equally rapid defensive measures. The reliance on open-source tools that can adapt to different AI models presents a unique challenge for organizations managing their identity and access controls.