AI Agents Breach More Than 100 Company Websites

Autonomous artificial intelligence agents have breached more than 100 company websites across the United States, Canada, Australia, and Japan, according to cybersecurity firm NorForge. The campaign, which began in late September 2026, targeted more than 1,200 websites and affected businesses including John Deere, SharkNinja, and Cummins, with the attacker relying on the open-source Cairn platform to coordinate operations.

NorForge's investigation found that approximately 10% of the attempted intrusions succeeded. The company estimated that the campaign cost its Chinese operator about $5,000 in AI usage fees, while successful intrusions required an average of just under 23 hours of agent work.

The findings illustrate how autonomous AI systems can coordinate complex cyberattacks with limited human involvement. NorForge also identified more than 20 exposed Cairn instances that researchers considered potentially malicious.

Snorre Fagerland, a senior researcher at NorForge, described the development in comments to Reuters.

"It's a new world," said Snorre Fagerland, senior researcher at NorForge.

The attacker reportedly used Cairn, an open-source problem-solving platform available on GitHub, to coordinate multiple AI agents around a defined objective. Investigators believe the system operated with Claude Code, Anthropic's coding tool, alongside DeepSeek V4.1 Flash, a lower-cost Chinese AI model.

According to NorForge, the operator supplied instructions in fewer than 100 Chinese characters. The agents were directed to find ways to execute code on thousands of websites without logging in, then install a small program that could upload files without administrators' knowledge.

The program functioned as a backdoor, potentially allowing persistent access to targeted systems even after the original vulnerability was patched. The operation sought continued access to payment systems associated with targeted online stores.

Researchers discovered exposed Cairn control panels beginning October 2, allowing them to observe the activity in real time without the operator's apparent knowledge. NorForge identified three entities using Cairn to coordinate large-scale AI-assisted cyberattacks, including the operator responsible for the campaign against more than 1,200 websites and an English-speaking hacker whose instructions included requests to obtain payment card information or cryptocurrency.

John Deere said it had taken corrective measures and found no evidence that sensitive information had been accessed. SharkNinja said the incident affected a single website dedicated to product support and that the threat had been contained quickly.

Cummins spokesperson Melinda Koski told The Verge that the company was investigating a possible cyberattack involving a website operated by one of its third-party service providers. The company did not confirm that sensitive information had been compromised.

Cairn's creator, who uses the pseudonym le1xia0, rejected the characterization of the platform as a hacking tool in a post on X. The developer said the software had not been designed for cyberattacks and argued that malicious use reflected the decisions of its users.

NorForge, however, characterized Cairn's role in this particular operation differently, describing the system as an AI-powered hacking platform in the context of the documented attacks. The researchers said they were working to remove exposed Cairn instances identified as malicious.

The investigation adds to concerns about the use of autonomous AI agents in cybercrime. NorForge's findings indicate that such systems can carry out repeated intrusion attempts, install persistent access mechanisms, and target numerous organizations with relatively little direct human supervision.