AI agents broke into more than 100 company websites at $50 each
The hacker gave instructions of less than 100 Chinese characters, and the agents did the remaining work.
Claimed, not confirmed
On 19 October 2026, researcher Snorre Fagerland says that a hacker who speaks Chinese used AI agents to break into websites. The hacker hit more than 100 companies, and Deere is one of them. The agents tried more than 1,200 websites. Fagerland estimates the cost at about $50 for each break in that gave access. The hacker did almost none of the work.
How it works
The hacker started on 25 September 2026. The agents then looked for weak points in each website for about 23 hours without help. The targets were e-commerce websites in the US, Canada, Australia and Japan.
Fagerland says the tools seem to be a mix of DeepSeek and Claude Code. The agents work on Cairn, an open-source platform for AI agents.
The numbers
Fagerland of NorForge estimates that the agents used about $5,000 of tokens (units that measure AI use). In one example, the agents found a control panel with a weak point. They used it to send bad commands to the server.
What is not known
The quantity of data that the hacker got is not known. Deere says it has no sign of access to sensitive data. Cummins says it looks into a possible attack on the website of a vendor.
The author of Cairn says it was not made for attacks.
Insight: Attacker errors are the main view that defenders have
Fagerland found about two dozen open Cairn systems that criminals used. In October 2026, open Claude Code records let CrowdStrike connect attacks on Korean banks to one hacker. This reads as a short time to see attacks: AI agents make attacks low cost, and attacker errors show them to defenders. If attackers close their tools, defenders must find attacks without that help.
Sources
Posted
