Hacker uses AI agents to breach over 100 company websites, Reuters reports

A Chinese-speaking hacker used artificial intelligence agents over the past few weeks to break into more than 100 company websites around the world, Yonhap and CBS NoCut News reported on 20 October, citing a Reuters report of 19 October (local time) that drew on a report by cybersecurity firm NorForge. Deere, Shark Ninja and Cummins were among the companies affected. [ 1 , 2 , 3 , 4 ]

NorForge senior researcher Snorre Fagerland wrote in the report that he found the hacker while searching the web for cases where Cairn, an open-source AI problem-solving platform popular with hackers, had been accidentally exposed on the internet. According to Fagerland, the hacker told the AI agents to investigate and break into target sites autonomously for hours, using instructions of fewer than 100 Chinese characters. [ 3 , 4 ]

Through last week the agents tried to break into more than 1,200 company websites and succeeded at more than 100, taking 23 hours on average to get in. Fagerland said the attack appeared to use a mix of China's DeepSeek and Anthropic's Claude Code and estimated the token cost at 5,000 dollars in total, or about 50 dollars per successful hack. He said that since about 2 October he had found 24 cases of Cairn being used for malicious activity. [ 3 , 4 ]

Deere told Reuters it had taken the necessary steps and that there was no sign the hacker had accessed sensitive information. Shark Ninja said the breach was limited to a standalone site holding only product support information and that the threat had been blocked. Cummins said it was investigating a report of a potential cyberattack on a website run by one of its external partners. Reuters could not confirm how much data was leaked, Yonhap said. [ 3 , 4 ]

Fagerland called the case a real-world example that commercial AI agents can carry out a complex intrusion from start to finish without human involvement, and said: "A whole new world has now opened." He had earlier expected hacking AI agents to need at least minimal human supervision, but said it was now clear the technology had moved past that stage. [ 3 , 4 ]

Cairn's creator wrote in a recent post on X that Cairn is an innovative form of multi-agent architecture and was not designed for hacking attacks, and that if it is used for harmful hacking, that is not his intention. Chosun Ilbo wrote that AI's ability to find security weaknesses itself and carry out complex attacks could put companies' information security on alert. [ 2 , 3 , 4 ]