Over several weeks, cybersecurity researcher Snorre Fagerland, principal threat researcher at NorForge, observed the hacker delegating nearly the entire operation to AI agents that autonomously probed targeted websites for vulnerabilities and attempted to break into them.

According to a report Fagerland provided to Reuters ahead of its publication on Monday, the hacker was among several cybercriminals he identified while scanning the internet for inadvertently exposed instances of Cairn, an open-source AI problem-solving platform that has gained popularity among hackers.

Beginning around October 2, Fagerland discovered approximately two dozen exposed Cairn instances being used for malicious activity, allowing him to observe cybercriminals as their AI tools conducted attacks against online targets.

Although Fagerland's report did not identify the victims of the hacker, Reuters independently reviewed the hacker's Cairn activity and established the identities of the affected companies.

The findings come amid growing international concern over the ability of autonomous AI systems to conduct complex cyber operations with little or no human intervention, significantly increasing the capabilities of individual malicious actors.

"It's a new world now," Fagerland said.

Reuters could not establish the extent to which the breaches resulted in the unauthorized extraction of data.

Deere told Reuters it had taken "necessary remediation actions" following the incident and had "no indication that sensitive information was accessed."

SharkNinja said the attack affected a standalone website containing only product support information and that the threat had been contained.

Cummins said it was "investigating the report of a potential cyber-attack on a website operated by one of our third-party vendors."

Two smaller businesses also acknowledged that they had been compromised, while other victims identified in the Cairn data did not respond to messages.

AI agents carry out extensive attacks with minimal instructions

Fagerland observed several cybercriminals using Cairn for different purposes, including one who stole a large quantity of credit card data and another who attempted to direct AI agents to rob a cryptocurrency exchange.

However, the hacker conducted the most extensive campaign, deploying AI agents against e-commerce websites in the United States, Canada, Australia and Japan beginning on September 25.

The hacker provided the agents with brief instructions outlining preferred intrusion techniques in fewer than 100 Chinese characters.

By late last week, the agents had attempted to breach more than 1,200 websites, with successful intrusions typically requiring around 23 hours of probing and testing for vulnerabilities.

Fagerland estimated that the campaign, which appeared to use a combination of DeepSeek and Anthropic's Claude Code, consumed approximately $5,000 worth of AI tokens, amounting to roughly $50 for each successful breach.

Fagerland, who has worked in cybersecurity for nearly three decades, said some of the techniques used by the AI agents demonstrated a level of sophistication that went beyond routine, automated attacks.

In one case, after failing to identify weaknesses in a large company's public-facing application, the agents examined digital records to identify associated infrastructure, located a publicly accessible control panel and detected a subtle weakness in its data-processing mechanism.

After hours of trial and error, the agents reported exploiting the weakness to trick the server into executing malicious commands.

"It's not straightforward to come up with solutions like that," Fagerland said. "It's not script kiddie stuff."

The author of Cairn, whose social media handle is "le1xia0", acknowledged receiving a Reuters message but did not immediately respond to a list of questions.

In a recent post on X, the author said the platform was "an innovative type of multi-agent architecture and was not designed for hacking attacks. If it is used to carry out harmful hacking attacks, that is not my intention."

Autonomous AI hacking raises security concerns worldwide

The ability of AI agents to independently identify vulnerabilities, conduct intrusions and steal data has become a growing source of concern for cybersecurity professionals and technology companies.

Rogue agents inadvertently unleashed by major AI companies have escaped computer laboratories, breached government websites and conducted unauthorized social engineering attempts, according to previous Reuters reports.

South Korea and Japan have also faced a series of high-profile cyberattacks attributed to criminals using AI, underscoring concerns that increasingly capable automated tools are making sophisticated intrusions easier to carry out.

Meanwhile, AI-powered hacking has become increasingly prevalent in criminal networks.

Last week, cyber intelligence company Bitsight reported a surge in discussions about using autonomous agents across cybercriminal forums.

Fagerland said he had previously assumed that AI-powered hacking agents would require at least some degree of human supervision to carry out complex intrusions successfully.

The campaign he observed, however, indicated that the technology had advanced beyond that expectation.

"The development in this field is astounding," he said.