A Chinese-speaking hacker used AI agents to breach over a hundred corporate websites. He worked with Cairn, an open-source platform that autonomously searches for vulnerabilities in applications and is popular among hackers. According to Reuters, the activity was highlighted by Snorre Fagerland, the chief threat researcher at NorForge.
How the Hacker with AI Agents Operated
Fagerland stumbled upon the trail while scanning the internet for unprotected instances of Cairn. Since October 2, he found about twenty instances being exploited for malicious activities, as reported by thenextweb.com. The Chinese-speaking attacker was the most active among those observed, giving instructions to his agents using fewer than a hundred Chinese characters.
The campaign started on September 25, 2026, targeting e-commerce sites in the United States, Canada, Australia, and Japan. In the week ending October 18, the agents attempted to infiltrate over 1,200 websites. A successful breach typically took around 23 hours of vulnerability testing. According to thenextweb.com, the operation combined the DeepSeek model with Claude Code from Anthropic.
Affected Companies: Deere, SharkNinja, and Cummins
Among those affected are agricultural equipment manufacturer Deere, home appliance maker SharkNinja, and engine producer Cummins. Deere stated, according to Reuters, that it has taken necessary remedial actions and has no indications that sensitive information was accessed. SharkNinja described the breach as an attack on a standalone website providing product support information and deemed the threat manageable. Cummins is investigating reports of a possible cyber attack on a website operated by one of its suppliers.
Cost of the Campaign
Fagerland estimated that the campaign consumed tokens worth about $5,000, or around $50 per successful breach. According to thenextweb.com, NorForge reports an average of about $5 per task, including failures. Typical targets were e-commerce sites with traffic ranging from 200,000 to 1.5 million, aiming for persistent access to business and payment systems.
In one case, the agents did not find a flaw in a public application but discovered a connected control panel, revealed a subtle vulnerability, and forced the server to execute malicious commands. "It's not easy to come up with such solutions. This is not the work of any script kiddie," Fagerland told Reuters.
Other hackers monitored by the researcher stole a batch of credit card data, while another attempted to rob a cryptocurrency exchange. The author of Cairn, known by the nickname "le1xia0," acknowledged the Reuters report according to cp24.com but did not immediately respond to questions. Two smaller businesses admitted they were also attacked.