Chinese Hackers Used AI Agents to Breach ConnectWise and Spy on Thousands of Companies
In the past week, the cybersecurity industry received what many analysts are calling a new milestone in the evolution of cyberattacks: a group of hackers linked to the Chinese state, known as Storm-0940, used artificial intelligence agents to compromise ConnectWise's cloud backup infrastructure and, from there, accessed Microsoft 365 environments in more than 2,250 organizations worldwide. The attack, detailed on October 21 by Microsoft Threat Intelligence, is notable not only for its scale — but for the speed with which it was executed. ARTEX agents carried out approximately 2,900 login attempts per minute, leaving defenders with a detection window of just minutes.
The vector: a digital supply chain
ConnectWise is one of the most important companies in the global IT ecosystem. Based in Florida, United States, the company develops IT management platforms for managed service providers (MSPs) — the third-party companies that administer systems for small and midsize businesses around the world. Its Cloud Backup product, formerly known as SkyKick, enables these providers to backup customer data in Microsoft 365 and Google Workspace environments.
The service works through a straightforward but powerful mechanism: when an MSP sets up Cloud Backup in a customer tenant, the application obtains permissions via Microsoft Entra ID to access emails, files, and Teams data on behalf of the customer. It is precisely this pre-established trust that the hackers exploited. Instead of trying to breach each company individually — which would have been slow, expensive, and easily detectable — they compromised ConnectWise's own identity credentials and began authenticating to customer tenants as if they were the legitimate application.
Starting on October 15, according to Microsoft, ARTEX agents began authenticating directly into customer Microsoft 365 environments using ConnectWise application secrets. Since the requests came from an automated script — without interactive user login — protections like multi-factor authentication (MFA) were completely ineffective. The system did not see an intruder; it saw authorized software doing what it was programmed to do.
ARTEX: the framework that changed the game
The most interesting — and disturbing — element of this incident is the use of ARTEX. It is a penetration testing framework based on AI agents, developed in Go and released as open source. In October, researchers linked ARTEX to attacks against South Korean banks, leading its developer to make it proprietary on October 8. But copies were already in circulation and continued to spread.
ARTEX allows the complete automation of a cyberattack cycle: network reconnaissance, password spraying, OAuth permission abuse, credential modification, and privilege escalation — all orchestrated by AI agents that can operate 24 hours a day, 7 days a week, without fatigue and without constant human supervision.
Microsoft observed that in some ARTEX-driven attacks, the time from initial targeting to privileged access was under three hours. The Microsoft research team's statement is direct: agents are not constrained by factors affecting humans — they can work around the clock, on weekends, and on holidays, and attempt multiple breach paths simultaneously, significantly reducing the time from target selection to objective achievement.
Storm-0940 was not the only group using ARTEX. Microsoft stated it had observed threat actors in several geographic regions using the same framework in their operations. What distinguishes the ConnectWise attack is not the technology in isolation, but the scale and sophistication of execution.
The timeline of a large-scale attack
Early indicators of the attack date back to mid-September 2026. Between October 10 and 14, Storm-0940 infrastructure at IP address 144.172.64.99 conducted an automated password spraying campaign against 327 accounts across eight organizations, generating more than 151,000 failed login events. At the time, most attempts were blocked.
Everything changed on October 15. With compromised ConnectWise Cloud Backup credentials, the AI agents changed strategy: they abandoned password spraying and began authenticating directly as legitimate applications. From there, the speed was devastating. By October 16, the Huntress detection platform had already identified at least 162 compromised tenants, and later reported nearly 1,000 tenants that received login attempts from malicious infrastructure, with more than 300 cases apparently successful.
What makes this attack particularly concerning is persistence. In one tenant analyzed by Huntress, researchers observed the attacker establishing permanent access in approximately five minutes: the MS Teams Cloud Management application self-granted Global Administrator role and added credentials to MSP management tools and to Huntress's own application.
Impact on organizations
ConnectWise confirmed that more than 2,250 Microsoft 365 tenants were targeted, though it has not disclosed the total number compromised. The company disabled seven application identities on October 17 and shut down the backup service. As of the time of reporting, Cloud Backup, SaaS Security, and Cloud Management remain suspended, causing significant operational disruption for thousands of companies that depend on these services for managing backups, security monitoring, and licensing.
The most affected sectors include defense, energy, financial services, government, legal, and manufacturing — sectors where data exposure can have consequences far beyond a corporate email breach.
ConnectWise stated that there is no indication that the backup data itself was compromised. The backup service itself remains encrypted and isolated. But access to active tenants — with the ability to read emails, modify files, and access Microsoft Teams — already represents an enormous risk.
What this means for IT security
The ConnectWise attack illustrates a fundamental shift in the nature of cyber threats. A few years ago, a supply chain attack required physical access, sophisticated social engineering, or extremely expensive zero-day vulnerabilities. Today, the combination of open-source AI tools, pre-established trust credentials, and the growing complexity of cloud infrastructures has created an environment where a single point of failure can open doors to hundreds of organizations.
According to Christian Bolander, a security analyst interviewed by CRN: "Years ago, when AI wasn't used the way it is today, you could make some mistakes and still get by on luck in many cases as an MSP. Now hackers are using AI and moving a thousand times faster!"
The question is no longer whether AI agents will be used in attacks — this is already happening. The question is whether defenses can evolve at the same speed, or whether we will enter a technological arms race where attackers are already running at machine speed and defenders still rely on human analysis to detect anomalies that, at scale, are indistinguishable from legitimate behavior.
Sources: TechCrunch, Microsoft Security Blog, The Hacker News, CRN, BleepingComputer
✓ Independent sources cross-checked and verified before publishing
