A hacker used artificial intelligence (AI) agents to hack into websites belonging to more than 100 companies, including the tractor manufacturer Deere, SharkNinja, and Cummins. The operation attracted attention for its ability to carry out complex attacks with little human intervention.
Cybersecurity researcher Snorre Fagerland of NorForge monitored some of the activity through the Cairn platform. According to his estimate, the campaign cost about US$5,000 (R$25,250) in AI credits.
The AI agents spent about 23 hours making continuous trial-and-error attempts before breaching the defenses. – Image: Mer_Studio/Shutterstock
More than 1,200 websites targeted
According to Reuters, cybersecurity researcher Snorre Fagerland discovered the operation while monitoring Cairn installations that were exposed on the internet and used for malicious activities. The open-source platform allows AI agents to be coordinated to perform tasks. Starting on October 2, the specialist managed to monitor the systems’ work and identify how they were being used in the attacks.
One of the cases that caught Fagerland’s attention was a Chinese-speaking hacker who, since September 25, had been directing the agents at websites in the United States, Canada, Australia, and Japan.The instructions were short, with fewer than 100 characters in Chinese, but were enough to guide the systems in their search for vulnerabilities and attempts to break into systems.
By the end of the week before the report was published, the agents had attempted to break into more than 1,200 websites. Successful intrusions generally took about 23 hours and cost an estimated US$50 (R$249) each. The operation apparently used a combination of DeepSeek with Claude Code, an Anthropic tool.
For Fagerland, the ability to carry out such complex attacks with little human intervention represents a significant shift in the cybersecurity landscape. “It’s a new world now,” Fagerland told Reuters.
How the agents found vulnerabilities
One of the incidents monitored by Fagerland helps explain how the AI agents carried out the intrusions. After failing to find flaws in a large company’s public-facing application, the systems changed strategy and analyzed digital logs to trace the infrastructure linked to the target. The search led to a control panel accessible over the internet, where they identified a flaw in data processing.
After hours of attempts, the agents said they had exploited the vulnerability to make the server execute malicious commands. The researcher was impressed by the ability to find an alternative path and test the flaw. “Finding solutions like these isn’t easy. It’s not the work of an amateur who merely runs scripts,” Fagerland said.
The case was not an isolated one.While monitoring other activity on the same platform, the researcher found different attempts at criminal use of AI: one hacker stole a large amount of credit card data, while another attempted to use agents to rob a cryptocurrency brokerage.
Specialists warn of a new era in cybercrime, marked by the low cost and automation of cyberattacks. – Image created by artificial intelligence
Companies investigate the incidents
In one of the attacks he monitored, Fagerland observed how the AI agents bypassed an obstacle during the intrusion. Since they found no flaws in the company’s public application, the systems began examining digital records in search of other ways to access the target’s infrastructure. The strategy led them to a control panel exposed on the internet, where they discovered a flaw in data processing.
Based on this discovery, the agents spent hours testing ways to exploit the vulnerability. Ultimately, they reported that they had induced the server to execute malicious commands. For Fagerland, the approach demonstrated a level of technical capability that went beyond running prewritten scripts. “Finding solutions like these is not easy. It is not something an amateur who merely runs scripts could do,” the researcher said.
An analysis of other Cairn installations revealed that this was not the only type of crime committed with the agents’ help. In one operation, a hacker stole a large amount of credit card data. In another, the criminal attempted to use AI to rob a cryptocurrency brokerage.The cases show that technology was being used at different stages of criminal activities, not just in the search for vulnerabilities on websites.
Valdir Antonelli is a journalist specializing in digital marketing and consumer behavior.
