One Click by Hackers, Rp90 Million Cost, 1,200 Websites Breached
Jakarta, CNBC Indonesia – Hackers needed only a brief instruction to deploy artificial intelligence (AI) agents and breach the websites of more than 100 companies. The attacks were carried out almost entirely automatically, with little human intervention.
The Mandarin-speaking hacker used AI agents to search for security vulnerabilities and infiltrate company websites over several hours. Among the victims were agricultural equipment manufacturer Deere, household appliance company SharkNinja and engine manufacturer Cummins.
Snorre Fagerland, a principal threat researcher at cybersecurity firm NorForge, uncovered the activity after finding an open-source AI platform called Cairn being used for hacking.
In his report, Fagerland said the hacker began the operation on 25 September. The hacker gave the agents brief instructions in Mandarin, fewer than 100 characters long, to direct them to carry out the attacks.
By last week, the agents had attempted to breach more than 1,200 websites in the United States, Canada, Australia and Japan. Successful intrusions generally took about 23 hours to identify and exploit a security vulnerability.
Fagerland estimated that the operation cost about US$5,000, or approximately Rp90 million, in AI tokens – around US$50 (Rp902,000) for each successful breach. The system is believed to have used a combination of China’s DeepSeek technology and Anthropic’s Claude Code.
In one case, an AI agent failed to find a weakness in a major company’s public-facing application. It then searched digital records for related infrastructure, identified an internet-accessible control panel, and exploited a flaw in data processing that allowed the server to run malicious commands.
“This is not something for which it is easy to find a solution,” Fagerland said, as quoted by Reuters on Thursday (22 October 2026). “This is not the work of amateur hackers,” he added.
He said he had previously expected AI agents used for hacking to require human supervision. But the findings showed the technology had advanced beyond his expectations.
“It’s a new world now,” Fagerland said.
Deere said it had taken the necessary remedial steps and found no indication that sensitive information had been accessed. SharkNinja said the attack targeted a separate website containing only product support information, and that the threat had been contained.
Cummins said it was investigating reports of a suspected cyberattack on a website operated by one of its third-party vendors.
Reuters could not independently establish the extent to which the attacks resulted in unauthorised data extraction.
The findings add to concerns about the capabilities of autonomous AI agents, which can identify vulnerabilities and carry out intrusions, potentially stealing data without requiring human direction at every stage.
“Developments in this field are truly astonishing,” Fagerland concluded.