One Hacker Let AI Agents Do the Hacking, and They Broke Into More Than 100 Companies

A cybersecurity researcher spent the past two weeks watching a hacker break into the websites of more than 100 companies, and the hacker did almost none of the work himself. The intrusions were carried out by AI agents, working on their own for hours at a time, Reuters reported on Monday.

The victims included farm equipment maker Deere, home appliance company SharkNinja and engine maker Cummins, Reuters said.

Watched through an exposed platform

Snorre Fagerland, principal threat researcher at the security company NorForge, found the campaign while scanning the web for exposed copies of Cairn, an open source AI problem solving platform popular with hackers. From Oct. 2, he found about two dozen copies of Cairn being used for malicious activity, which let him watch quietly as criminals set their AI tools loose, Reuters reported.

Fagerland shared his findings in a report given to Reuters ahead of publication on Monday. It did not name the victims. Reuters said it identified Deere, SharkNinja and Cummins by reviewing the hacker's Cairn activity itself.

"It's a new world now," Fagerland told Reuters. He said the case is real evidence that AI agents on sale today can carry out complex intrusions from start to finish without human guidance, a big jump in what one criminal can do alone.

Short orders, long attacks

The campaign began on Sept. 25, according to The Cryptonomist, which also covered the NorForge report. The hacker gave the agents their orders in fewer than 100 Chinese characters, setting out his preferred tactics. The agents then spent hours hunting for weaknesses and trying to exploit them.

By late last week the agents had targeted more than 1,200 websites, most of them online shops in the United States, Canada, Australia and Japan, and had breached more than 100 company sites. A successful break in took about 23 hours of probing and cost about 50 dollars in AI tokens. Fagerland put the whole campaign at roughly 5,000 dollars and said it appeared to mix models from China's DeepSeek and Anthropic's Claude Code, The Cryptonomist reported.

In one case, the agents could not find a weakness in a large company's public app. They searched records for linked infrastructure, found a control panel open to the internet and spotted a subtle flaw in how it handled data. After hours of trying, the agents used the flaw to make the server run malicious commands, The Cryptonomist reported.

Reuters could not establish how much data, if any, was taken.

Companies respond

Deere said it had taken the needed remediation steps and had no indication sensitive information was accessed. SharkNinja said the intrusion hit a standalone website that held only product support information and that it had contained the threat. Cummins said it was investigating a report of a possible attack on a website run by one of its third party vendors. Two smaller businesses also confirmed compromises, The Cryptonomist reported.

Fagerland said the Chinese speaking hacker was the busiest of several criminals he watched. One stole credit card data. Another tried to point AI agents at a crypto exchange to rob it, Reuters reported.

The author of Cairn, who posts under the handle le1xia0, wrote on X the platform was not built for hacking attacks and harmful use was not his intention, The Cryptonomist reported.

The case follows earlier warnings this year. In August, Gambit Security found Russian speaking extortionists using Cursor AI in breaches at seven organizations. In September it documented an agent driven campaign that hit 25 companies and stole at least 600,000 credit card records, The Cryptonomist reported.