The quick read: • A Chinese-speaking cybercriminal used AI agents to break into the websites of more than 100 companies, including John Deere, SharkNinja and Cummins, according to a NorForge report seen by Reuters (Oct 19). • The campaign began Sept 25 against online shops in the U.S., Canada, Japan and Australia. By late last week the agents had gone after more than 1,200 sites, per Reuters. • The agents ran on Cairn, an open-source AI agent platform. NorForge researcher Snorre Fagerland estimates the campaign appeared to use a mix of Anthropic's Claude Code and China's DeepSeek. • His cost estimate: about $5,000 in tokens, about $50 per successful hack. A successful intrusion typically took about 23 hours. Fagerland: "It's not script kiddie stuff." • Per NorForge, the agents aimed to plant a web shell for lasting access to shop and payment systems, per The Next Web. • Deere says it took "necessary remediation actions" and has "no indication that sensitive information was accessed." SharkNinja says a standalone product-support site was hit and the threat is contained. Cummins is investigating a report about a site run by a third-party vendor. Reuters could not determine how much data, if any, was taken. • For a small business: the targets were mid-sized online stores.
“One Hacker's AI Agents Broke Into More Than 100 Companies' Websites”
who’s making the noise
Reuters exclusive by Raphael Satter, Mon Oct 19, 2026, on a NorForge report by Snorre Fagerland. Campaign from Sept 25, 2026. Also reported by The Next Web.
our take
The number that matters is the price. By the researcher's estimate, a successful break-in cost this hacker about $50 in AI tokens, and the whole run about $5,000. That is the budget of a hobby, aimed at more than 1,200 online stores. The tools are not exotic. The agents ran on an open-source platform whose author says it was not designed for hacking, and NorForge estimates they appeared to draw on Anthropic's Claude Code and China's DeepSeek. Neither company is quoted in the reporting so far. The guardrails AI companies describe are built around a person typing a request. This was a fleet of agents working for about a day per target. The companies' answers are the usual first answers: remediation done, no sign sensitive data was accessed, a support site, contained, a vendor's site under review. Reuters could not establish how much was taken, so those answers may hold. They also show where the soft spots were: side sites and vendor sites, not the main storefront. The plain move for a small business: patch your store platform, remove admin accounts nobody uses, and ask your web vendor directly when they last checked the server for files that should not be there. What to watch: whether Anthropic or DeepSeek say anything about accounts tied to the campaign, and whether more victims are named.