Los AngelesNew YorkChicagoHoustonPhoenixPhiladelphiaSan FranciscoSeattleAtlantaBoston
Saturday, October 24, 2026Your daily read on the news.
Tech · AI

AI-Driven Breach Reaches Into Government and Defense Networks

A state-aligned actor is using autonomous agents to compromise cloud identities, threatening sensitive organizations.

By Mykola Bondarenko ·

A state-aligned actor has deployed autonomous AI agents in a large-scale attack that reached government and defense networks, according to a Microsoft report published on October 23. The activity, attributed to a group named Storm-0940, involves attempts to access more than 2,250 Microsoft 365 tenants. Microsoft's report indicates that in some instances, the group achieved privileged access within three hours of the initial targeting.

The operation centered on the ConnectWise supply chain, where unauthorized access to a multi-tenant application was observed. This breach provided stolen credentials for ConnectWise applications, allowing the attackers to authenticate without an interactive user sign-in. The report does not clarify how the credentials were obtained or how many tenants were ultimately compromised, leaving the full scope of the breach uncertain.

The speed and scale of the campaign were enabled by ARTEX, an open-source framework that runs a system of LLM-driven agents to perform an autonomous penetration test. This tool accelerated scanning, reconnaissance, and password spraying. The agents can operate continuously "with little needed human interaction," at a pace that "leaves defenders with little time to detect and respond."

The campaign involved a massive password spraying effort, generating more than 151,000 failed logon attempts against 327 accounts across eight organizations. Microsoft's data shows no successful logins from this method. On its peak day, October 11, the attack generated over 111,000 failed attempts, with almost 98% of the activity being locked out.

In contrast to the failed password attempts, the AI agents used the stolen application credentials to sign in to customer tenants. At one point, more than 200 sign-ins were completed in under four minutes using four different application credentials. The report states that the activity did not result from a compromise of or a vulnerability in Microsoft's systems, products, or services.

Targets for the campaign included industries such as aerospace and defense, semiconductor, government, and think tanks, with 14 listed in total. Storm-0940 has been active since 2021 and is described as a "China-state aligned actor" with "no known industry aliases." The name follows the convention for a developing actor.

Security firm Huntress observed unauthorized activity within tenant environments, including an application adding itself to a top administrator role. On October 15, Huntress saw an app perform this action, which corresponds to a Microsoft Defender alert titled "Application self-assigned Global Administrator role." The attackers also planted new secrets set to expire in 2028. By October 16, Huntress had confirmed unauthorized sign-ins for 162 tenants.

The use of ARTEX in this attack aligns with a broader pattern. Public reporting this month tied the tool to breaches at South Korean financial institutions. In those separate attacks, CrowdStrike said the suspect was likely based in China and allegedly used ARTEX with Anthropic's Claude Code. ARTEX's developer, who goes by the handle "Autumn-27," said on GitHub that the ARTEX project "will no longer be updated" and would become closed source, Reuters reported.

In response, ConnectWise has suspended several services as a precaution, including Cloud Backup (formerly SkyKick), SaaS Security, and Cloud Management. A phased restoration is underway, with recovery expected to take several days. The company noted it has no indicators to date that partner backup data was compromised. Microsoft advises customers to keep affected app connections disabled until formal restoration instructions are issued and to review activity from this month where logs allow.

Source: Tom's Hardware

More from AI