A Chinese-speaking hacker employed commercially available AI agents to breach over 100 organizations, including major companies such as John Deere, SharkNinja, and Cummins. The campaign, which began on September 25, involved attempts on more than 1,200 websites and incurred an estimated cost of $5,000 in tokens, or about $50 per successful intrusion. Snorre Fagerland, a principal threat researcher at NorForge, observed the attacks by monitoring inadvertently exposed instances of Cairn, an open-source AI problem-solving platform used by the hacker.
NorForge's investigation, detailed in its "Inside Cairn" report published on October 19, identified over 20 publicly exposed control panels for the Cairn platform. These panels offered a view into how threat actors selected targets, directed autonomous agents, exploited vulnerabilities, and established persistent access. Fagerland noted that the AI agents, reportedly using a combination of China's DeepSeek and Anthropic's Claude Code, conducted complex intrusions with minimal human intervention.
The attacks targeted e-commerce and payment infrastructure primarily in the US, Canada, Australia, and Japan, with the apparent goal of achieving persistent access for later monetization. NorForge's analysis of the exposed panels indicated financially motivated activity, with no clear signs of state-sponsored intrusion. The firm also discovered that some legitimate "red team" operations had left their Cairn panels exposed, potentially risking client data.
The companies affected have initiated responses to the breaches. John Deere stated it took "necessary remediation actions" and found no indication that sensitive information was accessed. SharkNinja reported that the hack affected a standalone product support site, and the threat had been contained. Cummins is investigating a potential attack on a website managed by a third-party vendor. Two smaller businesses also confirmed compromises.
Fagerland, with nearly three decades of cybersecurity experience, expressed surprise at the sophistication of some techniques used by the AI agents. In one instance, after failing to find weaknesses in a company's public application, the agents located associated infrastructure, identified a vulnerability in a control panel's data processing, and, after hours of attempts, exploited it to execute malicious commands. NorForge's findings highlight the evolving landscape of cyber threats, where autonomous AI tools can significantly amplify the capabilities of individual malicious actors.