Reuters and NorForge: One Operator Used AI Agents and About $5,000 in Tokens to Breach More Than 100 Company Sites
Reuters reported on Oct. 19 that a Chinese-speaking operator used the Cairn agent platform to attempt more than 1,200 sites and breach more than 100 companies' websites, including Deere, SharkNinja and Cummins. NorForge, whose research underpins the story, found more than 20 exposed malicious panels. Victim counts and costs come from panel data; Reuters could not establish whether data was taken.
Reuters on Monday, October 19, in an exclusive by Raphael Satter, reported that a Chinese-speaking hacker used Cairn, an open-source agent-orchestration platform, in a campaign that began on September 25. The operator's agents attempted more than 1,200 sites and broke into websites belonging to more than 100 companies, Reuters said, naming Deere, SharkNinja and Cummins. NorForge researcher Snorre Fagerland told Reuters the find was real-life evidence that commercially available AI agents could pull off complex intrusions from start to finish without human intervention.
What Reuters attributes to NorForge
Snorre Fagerland, principal threat researcher at NorForge, told Reuters that successful intrusions typically took around 23 hours and that the campaign — which appeared to use a mixture of China's DeepSeek and Anthropic's Claude Code — had used roughly $5,000 worth of tokens, or about $50 per successful hack. Reuters could not ascertain whether data was extracted. Deere told Reuters it had taken remediation actions and had "no indication that sensitive information was accessed"; SharkNinja said a standalone product-support site was hit and contained; Cummins said it was investigating a potential attack on a third-party vendor site. Cairn's author, "le1xia0", said on X that the platform "was not designed for hacking attacks."
What NorForge's report adds
NorForge's own report, published the same day, says it found more than 20 exposed malicious Cairn panels. One Chinese-speaking cluster it labels AIA0001 ran 14 panels and more than 1,100 projects aimed mainly at US, Japanese, Australian and Canadian e-commerce sites, with an average success rate around 10% and about 140 "facts" per project. Trufflehog scans found hundreds of secrets, 11% of them verified valid. Worker configurations were named claudecode_deepseek-v4.1-flash — a Claude Code harness driving DeepSeek V4.1 Flash, not Anthropic's models doing the hacking. NorForge separately estimates about $5 per task on average across successful and unsuccessful projects; that figure is not the same as Reuters' ~$50 per successful hack. A second, English-speaking cluster (AIA0002) focused on carding and crypto through an OpenRouter/DeepSeek setup.
"It's a new world now," Fagerland told Reuters.
Limits
- Victim counts, timing and costs come from exposed panel data as reported by Reuters and NorForge, not from independent incident reports. Reuters could not confirm data theft. Company statements are as Reuters relays them.