AI Agents Slash Corporate Attack Time to Under Three Hours
A new campaign uses autonomous software to breach thousands of business accounts, forcing major service providers to pause operations.
By Camila Rodriguez ·
Microsoft has linked a significant supply chain intrusion to a threat group known as Storm-0940, which is aligned with the Chinese state. The attack leveraged an open-source framework called ARTEX, which employs artificial intelligence to conduct penetration tests autonomously. This approach allowed the group to gain unauthorized access to a multi-tenant ConnectWise application and subsequently attempt sign-ins at over 2,250 Microsoft 365 tenants. In several instances, the attackers reached privileged access levels within less than three hours of the initial targeting.
The operation began with the theft of credentials belonging to ConnectWise applications. These credentials enabled the attackers to authenticate as the apps themselves, bypassing the need for interactive user sign-ins. Microsoft described ARTEX as a system of large language model driven agents that perform autonomous security testing. This technology accelerated scanning, reconnaissance, and password spraying. By operating around the clock with minimal human input, these agents left defenders with very little time to detect and respond to the intrusions.
Storm-0940 has been active since 2021 and does not have known industry aliases. The group targeted fourteen sectors, including aerospace and defense, semiconductors, and government think tanks. Microsoft clarified that the activity did not originate from a compromise or vulnerability in its own systems. A separate password spraying effort targeted 327 accounts across eight organizations. This campaign generated over 151,000 failed logons, with a peak rate of 2,900 attempts per minute on October 11. Nearly 98 percent of this activity resulted in account lockouts.
The identity takeover of ConnectWise cloud apps proved more effective than the password spraying. Using four application credentials, the AI agents signed in to customer tenants from the same IP range. They recorded more than 200 sign-ins in under four minutes. Within some tenants, the agents performed directory operations and reconnaissance. They also changed app credentials, identities, groups, and privileged roles. Huntress, a security firm, observed one tenant on October 15 where an app added itself to the Global Administrator role. This behavior triggered a specific alert in Microsoft Defender.
By October 16, Huntress confirmed unauthorized sign-ins at 162 tenants. The following day, the count rose to nearly 1,000 Entra tenants with sign-in activity from the attacking IP address, with more than 300 of those being successful. The attackers moved through tenants in sorted order, often seconds apart. This pattern suggested the use of a script working from a pre-defined target list. The use of a pentest framework that can work with any AI model distinguishes this campaign from many traditional attacks.
Other threat actors in several regions have used similar tools. Public reporting this month linked the framework to breaches at South Korean financial institutions. CrowdStrike stated that the suspect in those separate attacks was likely based in China and allegedly used the framework with Anthropic's Claude Code. Reuters reported that the developer of the framework, who goes by the handle Autumn-27, announced on GitHub that the project would no longer be updated and would become closed source.
ConnectWise detected the unauthorized access and suspended several services as a precaution. These included Cloud Backup, formerly known as SkyKick, along with SaaS Security and Cloud Management. The company has begun a phased restoration process expected to take several days. It currently reports no indicators that partner backup data was compromised. Microsoft cautioned that an affected app or a successful sign-in does not necessarily mean data was accessed or taken.
The company advised customers to keep affected app connections disabled until formal restoration instructions are issued. It also recommended that users review activity from this month where logs permit. This incident highlights the growing speed of AI driven attacks and the need for organizations to prepare for threats that can operate with minimal human oversight.
Source: Tom's Hardware
