China-linked hackers used AI pentest tool Artex in ConnectWise attacks

A state-sponsored Chinese hacking group used an AI-powered penetration testing tool called Artex in a campaign against ConnectWise, according to a SecurityWeek report published on 23 October 2026. The report said the attackers targeted hundreds of organisations and gained privileged access to some environments.

Penetration testing tools are built for defenders and hired testers: they probe systems for weaknesses the way an attacker would, so the owner can fix them first. AI-assisted versions automate more of that work, chaining together reconnaissance, exploitation and follow-up steps with less human direction. The same capability is useful to an intruder, and in this case, according to SecurityWeek, an offensive security tool was turned on real targets rather than consenting ones.

This is not a software flaw in the conventional sense. There is no reported bug in Artex itself. The problem described is tool misuse: legitimate automation pointed at systems its operators had no permission to touch. That distinction matters for defenders, because patching is not the remedy. Detection has to focus on behaviour, such as unexpected administrative activity or remote management sessions that do not match a known change window.

ConnectWise is widely used by managed service providers to administer customer systems, which is why compromise of such tooling tends to affect many organisations at once rather than one. SecurityWeek's report says privileged access was obtained in some environments, meaning the intruders held accounts or sessions with broad control, not just a foothold.

Details remain limited. The published report does not state when the activity began, how long it ran, which organisations were affected, or how the attackers first obtained access. It does not describe remediation steps or say whether affected customers have been notified. The Agentic Times has not independently verified the account, and ConnectWise has not been quoted in the material available.

The case adds to a pattern that security teams have been warning about for some time: the gap between an attack tool and a defensive tool is often only the intent of the person running it. Automated agents that can plan and execute multi-step intrusions shorten the time between initial access and privileged control, and they lower the skill needed to run a broad campaign against many targets at once.

Organisations that rely on remote management software may wish to review which accounts hold privileged access to it, how those sessions are logged, and whether anyone would notice an administrative action that nobody requested. Further details are expected as the reporting develops.