Skip to main content

Morning Overview

Microsoft says hackers are using an AI tool called ARTEX to power attacks worldwide

In mid-October, intruders holding stolen credentials for ConnectWise’s Cloud Backup service began hammering the cloud accounts of the companies that used it, at a pace of about 2,900 sign-in attempts per minute. Behind the speed, Microsoft says, was ARTEX, an open-source framework that strings together AI agents to run a penetration test from first reconnaissance to data theft with little human help.

Read next on Morning Overview: New FCC rules on how people can stop robocalls and texts take effect Nov. 25

Microsoft attributes the campaign to a China-aligned group it tracks as Storm-0940, and security firms now describe ARTEX-assisted intrusions against government, defense, finance and energy organizations in several countries. The episode is one of the clearest public cases yet of attackers folding agentic AI into an ordinary hacking operation.

ARTEX, an open-source multi-agent pentest framework

ARTEX was released earlier in 2026 on GitHub as an open-source project by an account called “Autumn-27,” The Record reported, and its developer was identified as Chinese cybersecurity engineer Li Puhua. Penetration testing tools are built for defenders, who hire testers to break into their own networks and find weaknesses first. ARTEX automates that job. According to an analysis by Sophos X-Ops, it is written in Go and uses planning agents to map a target and choose next steps while worker agents run offensive tools and feed the results back. It records discovered assets and vulnerabilities in a graph and can pursue several paths into a network at once.

The framework is model-agnostic. Sophos said operators in reported campaigns connected it to large language models including DeepSeek, GLM and Grok. On Oct. 8, Li posted a statement, machine-translated in The Record’s account, saying that given the misuse of the tool the ARTEX project would no longer be updated, and that it would become closed source. Sophos noted the original project has been taken offline and warned that copies on unofficial mirrors should be treated as untrusted; one backdoored ARTEX Docker image is already detected as malware.

Storm-0940 and the ConnectWise Cloud Backup breach

In a blog post dated Oct. 21, 2026, Microsoft linked the ConnectWise attacks to Storm-0940, a group it describes as Chinese government-aligned, active since 2021 and focused on data theft. The attackers used stolen Cloud Backup credentials to reach systems where the app was installed, a supply-chain route that turned a backup service protecting Microsoft 365 and Google Workspace data into a door. ConnectWise shut down several services the weekend before Microsoft’s disclosure.

The volume was extraordinary. More than 111,000 failed log-on attempts were recorded on Oct. 11 alone, 48 accounts each received more than 1,000 failures, and account lockouts made up nearly 98 percent of the activity, according to The Record’s summary of Microsoft’s findings. Most attempts failed. But “Storm-0940 achieved privileged cloud persistence at many organizations in a brief time,” Microsoft wrote, and in several cases it took under three hours to go from initial targeting to privileged access. Redmond magazine reported that the AI tooling was what compressed that timeline.

Victims in Asia, Europe and the United States

ConnectWise was one entry point among many. Since September, Storm-0940 has breached organizations in government, aerospace, defense, finance, manufacturing, technology and energy, Microsoft said, and The Record described the group as targeting hundreds of organizations globally, though no verified count has been published. Zscaler, in its own report on the campaign, found ARTEX-backed attacks against dozens of industries across multiple countries. ARTEX had already been tied to attacks on financial institutions in South Korea and Japan, and Sophos cited a CrowdStrike report of a September 2026 ARTEX-related incident aimed at South Korean financial firms.

Microsoft framed the significance in one line: “This activity demonstrates how threat actors are integrating agentic AI into existing offensive security workflows.” Chinese Foreign Ministry spokesperson Mao Ning denied knowledge of the incidents. GovInfoSecurity’s coverage likewise identified the attackers as China-aligned.

Same techniques, more noise and persistence

Defenders reading the reports may notice that nothing ARTEX does is new. Sophos said the underlying activities, such as webshells, SQL Server command execution, credential theft and lateral movement, look broadly the same as before, but “the scale, noise, and persistence all increase significantly.” The company recorded hundreds of protection events related to the attacks at a small number of customers. Operators fed ARTEX authentication material or compromised application credentials and turned it on cloud APIs such as Microsoft Graph, according to Microsoft and Google’s Mandiant.

The recommended defenses are correspondingly familiar. Sophos advises auditing Entra ID service principals, applying conditional access rules to workload identities, hunting back to at least September 2026 for signs of Storm-0940, and automating the first response to privileged access so that a three-hour attack is not met with a next-day review. Dark Reading characterized ARTEX as a pen-test tool turned into a weapon, and that is the open problem: the framework’s code was public before its author pulled it, and clones remain in circulation.

This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.


More from Morning Overview

Morning Overview is reader-supported. Some links in our articles are affiliate links, and we may earn a commission at no extra cost to you. As an Amazon Associate I earn from qualifying purchases. Full disclosure.