Autonomous AI Agents Breach Over 100 Corporate Websites

The attacker leveraged Cairn, an open-source platform designed as a general-purpose problem-solving engine. By providing fewer than 100 Chinese characters as instructions, the hacker prompted the AI agents to execute successful breaches in roughly 23 hours. Norforge identified over 20 exposed instances of the platform currently being used for what researchers classify as likely malicious activity. The attacks spanned infrastructure in the United States, Canada, Australia, and Japan.

Corporate responses to the discovery vary. John Deere stated there is no evidence that sensitive information was compromised, while SharkNinja confirmed the breach affected a standalone product support site and has since been contained. Cummins is currently investigating the incident, noting the potential involvement of a third-party vendor. Cairn’s creator, who operates under the handle @le1xia0, maintains that the platform was never intended as a hacking tool, arguing that misuse remains a human decision rather than a built-in feature of the engine. Norforge is now working to dismantle the malicious instances.