The hard work was done by the agents. They searched for flaws, tested, tried again and reached their goal in a little under a day on average. According to the researcher's estimate, each successful break-in cost about $50.
The case was discovered by the cybersecurity firm NorForge, which published a report on it on 19 October. The news agency Reuters identified, among the victims, websites linked to Deere, SharkNinja and Cummins.
The attackers were caught because of their own carelessness. In early October, NorForge's principal threat researcher Snorre Fagerland was searching the internet for exposed control panels of AI tools and also found the hacker's panels — unprotected. This let the researchers see exactly what the agents did. The attack had begun on 25 September.
What is an AI agent?
An ordinary chatbot answers questions. An AI agent takes steps on its own: it opens pages, runs commands, looks at the result and decides what to do next. Such agents are used, for example, for programming and for automating repetitive work. Unfortunately, the same capability also suits attacks.
How the cyberattack worked
The hacker used the open-source platform Cairn. It is given a target and a goal. The agents then move forward step by step and share their findings on a common board that all agents read. A human only sets the targets and occasionally adds hints.
According to Reuters, the hacker's instructions were shorter than 100 Chinese characters. In one case, the agents found no flaw in a large company's web application. They then found a linked control panel and, after hours of trial and error, got the server to run commands.
According to NorForge, the name of the agents' worker process suggests that the programming agent Claude Code was used together with the low-cost language model V4.1 Flash from the Chinese company DeepSeek. This is the researchers' interpretation, not a confirmed fact.
