AI agents autonomously broke into 100+ company websites for roughly $50 per successful hack

A Chinese-speaking hacker set AI agents loose on e-commerce sites across the US, Canada, Australia, and Japan starting September 25, issuing instructions in fewer than 100 Chinese characters and then largely stepping aside. The agents spent an average of 23 hours per target probing for weaknesses, attempted intrusions at more than 1,200 sites, and successfully breached over 100, including John Deere, SharkNinja, and Cummins. Total estimated cost: around $5,000 in API tokens, or about $50 per confirmed breach. The campaign appeared to run on a mix of DeepSeek and Anthropic's Claude Code, routed through Cairn, an open-source AI problem-solving platform that cybersecurity researcher Snorre Fagerland found exposed on the open web.

Fagerland observed the operation in real time after stumbling onto inadvertently exposed Cairn instances while scanning the internet. In one case, after failing to find weaknesses in a target's public-facing app, the agents independently hunted associated infrastructure, found a control panel, identified a subtle data-processing flaw, and exploited it via command injection — all without human input. Fagerland, who has worked in cybersecurity for nearly three decades, noted the agents' techniques were not script-kiddie territory. He had previously assumed hacker-agents would need at least some supervision. He no longer does.