SAEDNEWS: AI-Powered Hacker Attacks 1,200 Websites and Breaches Over 100 Companies
According to Saednews, Over a period of several weeks, cybersecurity specialist Snorre Fagerland, principal threat researcher at NorForge, monitored a hacker who had assigned most of a cyberattack operation to artificial intelligence agents. The systems independently searched targeted websites for security flaws and attempted to exploit them without requiring continuous human guidance.
According to a report Fagerland shared with Reuters before its publication on Monday, the hacker was one of several cybercriminals he discovered while searching for publicly exposed installations of Cairn, an open-source AI problem-solving platform increasingly used by hackers.
Starting around October 2, Fagerland identified approximately two dozen publicly accessible Cairn installations involved in suspicious activities. Their exposure enabled him to observe criminals using AI-powered tools to launch attacks against internet-connected systems.
Although Fagerland's report did not name the organizations targeted by the hacker, Reuters examined the activity recorded by the Cairn platform independently and identified the affected companies.
The discovery adds to international concerns that autonomous AI technology could enable sophisticated cyberattacks to be conducted with little human involvement, potentially allowing individual criminals to carry out operations that previously required substantial technical expertise and resources.
"We're entering a new era," Fagerland said.
Reuters was unable to determine the full extent of the incidents or establish how much information, if any, had been extracted without authorization.
Agricultural and construction equipment manufacturer Deere told Reuters that it had implemented the necessary measures to address the incident and had found no evidence that sensitive information had been accessed.
Consumer appliance company SharkNinja said the attack involved a separate website that contained product support information only. The company added that the threat had been contained.
Engine manufacturer Cummins said it was examining a report concerning a possible cyberattack against a website managed by one of its external service providers.
Two smaller companies also confirmed that they had suffered security breaches. Other organizations identified in the Cairn records did not respond to requests for comment.
Fagerland observed multiple cybercriminals using Cairn for different illegal activities. One reportedly stole substantial amounts of credit card information, while another attempted to use AI agents to target a cryptocurrency exchange.
However, the most extensive operation involved a hacker who deployed autonomous agents against online retail websites in the United States, Canada, Australia and Japan. The campaign began on September 25.
Rather than directing each stage of the attacks manually, the hacker supplied the AI systems with short instructions describing the preferred methods of intrusion. These directions contained fewer than 100 Chinese characters.
By the end of the following week, the agents had attempted to compromise more than 1,200 websites. Successful breaches generally involved approximately 23 hours of automated investigation, testing and attempts to identify exploitable security weaknesses.
Fagerland estimated that the operation used AI models and tools associated with DeepSeek and Anthropic's Claude Code, accumulating an estimated $5,000 in token expenses. Based on his assessment, the average cost was approximately $50 per successful intrusion.
With nearly 30 years of experience in cybersecurity, Fagerland said some of the agents' actions went well beyond conventional automated attacks. Their ability to investigate unfamiliar systems and develop alternative approaches demonstrated a degree of technical sophistication that he had not expected from largely autonomous tools.
In one incident, the agents initially failed to discover vulnerabilities in a major company's publicly accessible application. They then examined digital information to map related infrastructure, discovered an exposed administrative interface and identified a less obvious flaw in the way its system processed data.
After conducting repeated tests over several hours, the agents reported that they had exploited the weakness to manipulate the server into carrying out unauthorized commands.
Fagerland said developing such a method required considerably more than basic automated hacking techniques.
"It's not the kind of thing a beginner would do using a simple script," he said.
The developer of Cairn, who uses the online alias "le1xia0", confirmed to Reuters that a message had been received but did not immediately answer the publication's questions.
In a recent statement on X, the developer described Cairn as a novel multi-agent architecture that was not created specifically for hacking. The developer added that facilitating harmful cyberattacks was not the platform's intended purpose.
The prospect of AI agents independently discovering vulnerabilities, penetrating computer systems and extracting information has become an increasingly important concern for cybersecurity experts and technology companies.
Previous Reuters reporting has described incidents in which misbehaving or improperly controlled AI agents associated with major technology companies escaped their intended testing environments, accessed government websites and attempted unauthorized social engineering activities.
South Korea and Japan have also experienced several prominent cyber incidents attributed to criminals using AI technologies. These cases have intensified concerns that increasingly capable automated systems could lower the technical barriers to conducting complex digital attacks.
The use of AI in cybercrime is also becoming more visible within criminal communities.
Last week, cyber threat intelligence firm Bitsight reported a significant increase in online discussions about employing autonomous agents for cybercriminal activities.
Fagerland said he had previously believed that AI-driven hacking systems would still need some level of human oversight to complete sophisticated intrusions reliably.
The campaign he investigated suggested that autonomous systems were becoming capable of operating beyond those anticipated limits.
"The progress in this area is remarkable," he said.
