Hackers abuse AI pen-test tool Cairn in mass website attacks

A Chinese-speaking attacker has used Cairn, an AI-powered penetration-testing tool, to automate attacks against hundreds of websites around the world, Dark Reading reported on 20 October 2026.

According to the report, the campaign compromised hundreds of sites globally and resulted in data being taken from large companies, among them the agricultural machinery maker John Deere, the appliance firm SharkNinja and the engine manufacturer Cummins. Dark Reading described the activity as a "massive attack spree".

The case is an example of what security researchers call tool misuse: software built for legitimate defensive work being turned to offensive ends. Penetration-testing tools are designed to find weaknesses in systems with the owner's permission. When those tools add AI-driven automation, the same capability can be pointed at targets that never consented, and at far more of them at once. In this instance the attacker did not need to break Cairn itself; the reported problem is how it was used.

Several important details are not established by the available reporting. It is not clear when the intrusions began or how long they ran before being noticed, what kinds of data were taken from the named companies, or how the attacker obtained access to Cairn. The report does not describe which vulnerabilities or techniques the tool was directed to exploit, nor whether the affected organisations have confirmed losses. None of the named companies is quoted in the available material, and there is no statement from the makers of Cairn.

The incident is currently at the reported stage. Readers should treat the scale figure of "hundreds" of sites as the reporter's characterisation rather than a confirmed count from the affected organisations or from law enforcement.

For security teams, the practical significance is less about one particular tool and more about the category. Automated offensive tooling compresses the time between a weakness appearing and it being found by someone hostile. Defenders who rely on the gap between public disclosure of a flaw and manual exploitation have less room than they did. Routine controls apply: knowing what internet-facing assets you own, patching them promptly, watching for unusual scanning and authentication patterns, and logging enough to reconstruct what an automated campaign touched.

The wider question raised by the report is governance of dual-use AI tooling. Products that can test a network can also attack one. Vendors of such tools typically gate access behind customer verification, but the report does not say what controls Cairn had in place or whether they were bypassed, misused by an authorised user, or simply absent.

The Agentic Times has not independently verified the claims in the report. We will update this story if the named companies, the tool's developers or investigators publish further detail.