An attacker used AI agents to break into more than 100 websites, with little direct involvement from the human operator, according to findings from AI security firm Norforge. Reuters reported that the affected organizations included John Deere, SharkNinja, and Cummins.
The activity offers a concrete example of how a general-purpose AI platform can be directed toward cyberattacks. Norforge says it found exposed instances of the platform being used maliciously, and is working to get them taken down.
How the Cairn agents were used
Norforge uncovered the activity by examining exposed work on Cairn, an open-source platform that describes itself as a “general-purpose problem-solving engine.” Users provide the agents with a starting point and a goal, and the system records their progress as they work through a task.
That setup can also be used to coordinate an intrusion. In the case described by Norforge, a hacker gave agents a brief set of “preferred intrusion tactics” in fewer than 100 Chinese characters. The agents then carried out attacks with limited human involvement.
Reuters reported that successful breaches took about 23 hours. Norforge identified the operator as the “most active” among the exposed Cairn instances it found, and said the attacks targeted companies in the US, Canada, Australia, and Japan.
What the companies said about the breaches
The reported effects varied by company. Cummins spokesperson Melinda Koski told The Verge that the company “is investigating the report of a potential cyber-attack on a website operated by one of our third-party vendors.”
A John Deere spokesperson told Reuters there was “no indication that sensitive information was accessed” as a result of the attack. SharkNinja said the incident affected a standalone website containing product support information and that it had contained the threat.
Those statements describe the companies’ accounts of the incidents; they do not establish that every affected site or organization experienced the same kind of access or impact. The source report does not provide further details about what the attacker did on each site.
Exposed instances raise wider security concerns
Norforge found more than 20 exposed Cairn instances it considers “likely malicious,” and says it is working to get them taken down. The report points to a practical risk: agents designed to move toward a goal can be given harmful goals, too, and may perform much of the work with limited oversight.
The findings also leave important limits. The source does not name all of the affected websites, and Norforge’s findings on exposed instances do not by themselves describe the full scope of each breach. The public statements from the named companies offer only some information about the impact.
The platform’s creator says misuse is a human choice
Cairn’s creator, who goes by @le1xia0, said earlier this month that the platform “was never a ‘hacking tool,’” and added that “misuse is a human decision, not a feature of the engine.” That view distinguishes a general-purpose system from the way an operator chooses to use it.
For organizations, the reports underline that AI agents can be part of an intrusion workflow, even when the system was built for broader problem-solving tasks. Norforge’s effort to remove malicious exposed instances is one response; the reported attacks also show why the security of tools that can act autonomously matters alongside the intent of their users.