AI Agents Used in Automated Cyberattacks on Major Industrial Firms

Researchers at Norforge have detailed a large-scale intrusion campaign powered by autonomous AI agents operating on the open-source Cairn platform. The attacker defined preferred intrusion tactics in fewer than 100 Chinese characters, then let the agents execute the breaches with minimal human oversight. Successful compromises averaged roughly 23 hours each, hitting targets across the United States, Canada, Australia, and Japan.

Among the confirmed victims are John Deere, SharkNinja, and Cummins. John Deere stated there is no indication sensitive data was accessed, while SharkNinja said the breach was limited to a standalone product-support site and has been contained. Cummins is investigating a potential attack on a third-party vendor website. Norforge identified more than 20 exposed Cairn instances likely used for malicious purposes and is working to have them taken down.

Why it matters for GPU and AI infrastructure

This incident underscores how readily available GPU-accelerated AI agents can be repurposed for offensive security operations. The same compute clusters that train and serve large language models can also host swarms of autonomous agents capable of scanning, exploiting, and persisting across hundreds of endpoints. For GPU cloud providers and enterprise AI teams, the event reinforces the need for strict workload isolation, continuous monitoring of agent activity, and robust authentication on any platform that allows unrestricted code execution.

Cairn's creator maintains the platform was never intended as a hacking tool and that misuse is a human decision. Nevertheless, the episode demonstrates that open agent frameworks lower the barrier for automated attacks, making it essential for infrastructure operators to implement guardrails such as egress controls, runtime attestation, and audit trails for every agent deployment.