The control systems directing automated cyberattacks against corporate websites were themselves accessible over the public internet, allowing researchers to observe an operation that reportedly penetrated more than 100 websites and revealing a separate security weakness in AI-powered hacking infrastructure.
Cybersecurity researcher Snorre Fagerland of NorForge discovered publicly exposed instances of Cairn, an open-source platform used to coordinate autonomous AI agents, beginning around October 2, according to a Reuters investigation.
Fagerland identified approximately two dozen instances associated with malicious activity. Their exposed interfaces allowed him to examine ongoing operations without compromising the attackers’ systems.
Sponsored · Altamira Gold Corp.
One Chinese-speaking operator had instructed AI agents to attack more than 1,200 e-commerce websites across the US, Canada, Australia, and Japan, beginning September 25.
NorForge reported successful intrusions into more than 100 websites. Reuters independently identified affected websites associated with Deere, SharkNinja, and Cummins.
The findings draw attention to the infrastructure managing autonomous cyber operations, rather than only the vulnerabilities exploited by the agents. Cairn coordinates multi-step tasks and maintains records of an agent’s progress. When its management interface is publicly accessible, operational information can become visible to third parties.
A separate investigation by ReliaQuest documented another attack involving Cairn. Researchers identified an unauthenticated control panel operating on infrastructure used to compromise a server, involving hundreds of commands executed within 24 hours and resulting in the attacker obtaining administrative access.
The affected companies described differing consequences when contacted by Reuters. Deere said it had undertaken remediation and found no indication that sensitive information was accessed.
SharkNinja said the affected website was a standalone product-support service and that the threat had been contained, while Cummins said it was investigating a possible cyberattack involving a third-party vendor’s website.
According to NorForge, the operator’s agents typically required approximately 23 hours to complete a successful intrusion. Fagerland estimated total AI usage at roughly $5,000, equivalent to approximately $50 per successful compromise.
AI-assisted hacking was already documented before the latest disclosure. In September, Gambit Security reported a separate campaign involving at least 27 compromised companies and more than 600,000 stolen payment-card records.
