Hacker Used AI Agents to Breach 100+ Company Websites
· Cortex Solutions
Reuters and NorForge reveal a hacker who used autonomous AI agents to break into 100+ company websites for about $5,000. What happened and how Saudi firms can stay safe.
In an exclusive report published by Reuters on October 19, 2026, it emerged that a Chinese-speaking hacker managed to break into the websites of more than 100 companies over roughly two weeks, after handing most of the work to AI agents that spent long hours probing and breaking into targets with little human supervision. The story is based on a report published the same day by cybersecurity firm NorForge, written by principal threat researcher Snorre Fagerland, and was also covered that day by the tech outlet TNW.
How was the campaign discovered?
According to Reuters and TNW, Fagerland was scanning the internet for accidentally exposed instances of a platform called Cairn, an open-source system for solving problems with multiple AI agents. The platform takes a target and a goal, then splits the work among agents that advance step by step, posting their findings on a shared board that all of them can read, while a human operator can add hints.
From around October 2, the researcher found about twenty exposed control panels he judged likely to be used for malicious purposes, and grouped them into three clusters. The largest was a Chinese-language cluster with 14 panels and more than 1,100 attack projects.
What exactly did the agents do?
As reported by Reuters:
- The campaign began on September 25 and targeted online stores in the US, Canada, Australia and Japan.
- The operator wrote the instructions in fewer than 100 Chinese characters, then left the agents to work on their own.
- By late last week, the agents had attempted to break into more than 1,200 sites.
- Successful intrusions took around 23 hours on average.
- Fagerland estimated the campaign cost about $5,000 in model tokens, roughly $50 per successful hack. TNW stresses this is a rough figure, and that NorForge's report puts the average cost per task at about $5 when failed attempts are counted.
TNW says the apparent goal was to run commands on store servers and plant an uploader web shell to gain lasting access to shop and payment systems. In one case the agents found no flaw in a large company's main application, so they searched digital records for related infrastructure, found a public control panel, and exploited a subtle data-processing flaw until they tricked the server into running malicious commands.
As for tooling, the reports say the campaign appeared to use a mix of a DeepSeek model and Anthropic's Claude Code, as one worker's name suggested, but NorForge did not confirm the full setup.
Who were the victims?
Reuters identified Deere, SharkNinja and Cummins among the victims:
- Deere said it took "necessary remediation actions" and had no indication that sensitive information was accessed.
- SharkNinja said the hack hit a standalone product-support site and that the threat was contained.
- Cummins said it was investigating a potential attack on a website run by a third-party vendor.
Reuters could not determine how much data, if any, was taken. The researcher also observed other hackers: one stole a large trove of credit card data, and another tried to use agents to rob a cryptocurrency exchange.
A striking detail: exposed secrets
According to TNW, the exposed panel data contained hundreds of credentials, API keys and other secrets, and the Trufflehog tool verified 11% of them as still valid. Even more surprising, some legitimate penetration-testing firms had left their clients' work exposed in the same way.
Fagerland told Reuters he had assumed hacking agents would still need supervision, but the technology had clearly moved past that, calling the development "astounding" and saying the attacks were "not script kiddie stuff." Cairn's developer, known as le1xia0, said in a post on X that the platform was "not designed for hacking attacks."
What does this mean for Saudi businesses?
The key message is that the cost of an attack has dropped sharply, and an attacker no longer needs a professional team to try hundreds of targets at once. Online stores and mid-sized company websites were the preferred targets in this particular campaign. We therefore advise business owners in the Kingdom to:
- Review everything of yours that is exposed to the internet, not just the main website: control panels, subdomains, support pages, and systems run by outside vendors.
- Never leave the control panels of AI or automation tools publicly reachable; put them behind strong sign-in or a private network.
- Rotate API keys and passwords regularly, and never store them in code or shared documents.
- Keep store platforms and plugins updated, and turn on monitoring and alerts for any unfamiliar files or commands on the server.
- Ask your vendors about their security practices, because a breach of a site run by a third party still hurts your reputation.
At Cortex Solutions we build systems and applications with security in mind from day one, and we help companies adopt AI automation safely and under control. Learn more about our services: https://cortex.sa/services/ai-automation