adtestbench Higgsfield

AI news · Security · AI agents

Hacker used AI agents to breach over 100 websites, Reuters says

A security firm found the attacker’s control panels open on the internet, showing agents that probed targets, found weak spots and broke in with little human input.

By Alexandre S. , 00:30 UTC

One attacker, writing in Chinese, used a crowd of AI agents to get into more than 100 websites, Reuters reported on Monday 19 October 2026, drawing on research by security firm NorForge. Reuters, which helped NorForge alert victims, named SharkNinja, Cummins and John Deere among the companies hit, The Verge reports.

Retro-futurist illustration: a striped sunset over a grid horizon under a starry sky, with a shield motif standing on the horizon.
Drawn by adtestbench from “Inside Cairn: Open windows into massive AI-driven cyber operations”,

The agents ran on Cairn, a free tool that moves a team of AI workers step by step from a starting point to a set goal. Here the goal was access to a site. NorForge found more than 20 Cairn control panels left unprotected online that it links to criminal activity. The busiest group ran over 1,100 attack projects across 14 panels likely tied to it, all on a worker named after DeepSeek’s V4.1 Flash model.

The attacker’s instructions were brief, under 100 Chinese characters in all, Tom’s Hardware reports, yet the agents had gone after more than 1,200 sites by 18 October. NorForge estimates the average task cost about $5 in AI fees and calls the figure highly uncertain.

SharkNinja says only a standalone help site was affected and the threat is contained. Deere says it has found nothing to suggest sensitive data was touched, and Cummins says it is looking into a report about a site run by one of its suppliers.

For any brand with a website, an attack priced at a few dollars a try means even small sites are worth an attacker’s time.