Hackers Used AI Agents to Break Into Thousands of Microsoft Accounts

If you work with Microsoft 365, a new report suggests you should be aware of a sophisticated cyberattack that has been quietly probing accounts. According to a security blog post from Microsoft, a China-state-aligned actor identified as Storm-0940 has been using an AI-powered tool to attempt logins across more than 2,250 Microsoft 365 tenants. The attackers leveraged stolen credentials associated with ConnectWise applications to carry out these attempts. In some cases, the AI agents were able to gain privileged access to a system in under three hours after an initial attempt was made.

This isn't a simple guessing game. The attackers also engaged in automated password guessing, a technique known as password spraying, against 327 accounts belonging to eight different organizations. This resulted in over 151,000 failed login attempts but yielded no successful entries. However, a different part of the operation showed alarming efficiency. The group used ARTEX, an open-source framework for penetration testing that is driven by large language models (LLMs), to automate parts of the attack.

This AI-driven approach allowed the attackers to operate with a speed and autonomy that can be difficult for security teams to counter. Microsoft explained that the system consists of 'a system of LLM-driven agents that run an autonomous penetration test' and that these agents can work continuously 'with little needed human interaction.' This pace, Microsoft noted, 'leaves defenders with little time to detect and respond.' The framework was used to accelerate scanning, reconnaissance, and password spraying, as well as to abuse OAuth applications.

The timeline of this activity shows a coordinated and rapid escalation. The attack was not a single event but a sustained campaign. On October 11th, the password spraying activity reached a peak of 2,900 attempts per minute, all of which failed. In a separate part of the operation, on October 15th, a security firm named Huntress observed an unauthorized application adding itself to a tenant's top administrator role. By October 17th, nearly 1,000 tenants had shown sign-in activity from the attackers' IP address, with more than 300 of those sign-ins succeeding.

The activity has been linked to several sectors, including aerospace, defense, semiconductors, government, and think tanks, among a total of 14 listed industries. Microsoft clarified that this specific campaign did not result from a vulnerability or compromise within Microsoft's own systems, products, or services. Instead, it appears to stem from an unauthorized access to a multi-tenant application belonging to ConnectWise. The attackers used credentials stolen from that application to sign into customer tenants. It is currently unclear how those credentials were obtained or how many tenants were ultimately compromised.

The use of AI to carry out such an attack is a notable development. The ARTEX framework can use any AI model to perform its tasks. This technology is not being used by just one group. Public reporting this month linked similar activity to breaches at South Korean financial institutions. Security firm CrowdStrike stated that the suspect in those attacks was likely based in China and allegedly used ARTEX with Anthropic’s Claude Code. The developer of the ARTEX project, who goes by the handle ‘Autumn-27,’ has stated that the project will no longer be updated and will become closed source.

In response to the unauthorized access, ConnectWise has suspended several of its services as a precaution, including Cloud Backup, SaaS Security, and Cloud Management. The company has begun a phased restoration process and expects it to take several days. They have noted that they have no indicators to date that partner backup data was compromised. Microsoft advises customers to keep any connections to affected apps disabled until formal restoration instructions are issued and to review activity logs from this month where possible.