A Chinese-speaking hacker activated artificial intelligence agents to autonomously carry out cyberattacks against more than 1,200 websites, including those of major U.S. and international companies.
Të lidhura
None found
Details were made public in a report by cybersecurity researcher Snorre Fagerland. According to him, the hacker left nearly the entire process to the AI agents, which spent hours searching for vulnerabilities and trying to breach the systems of various companies.
The campaign’s targets included Deere, a manufacturer of agricultural equipment; SharkNinja, a home appliance company; and engine manufacturer Cummins.
Fagerland, a lead threat researcher at cybersecurity company NorForge, described the case as concrete evidence that AI can be used to carry out complex cyberattacks from start to finish, with minimal or no human involvement. “It’s a new world now,” he said.
A broad campaign with minimal instructions
The campaign began on September 25 and focused on attacking e-commerce websites in the United States, Canada, Australia and Japan.
The hacker sent the AI agents instructions in Chinese, each containing fewer than 100 characters. Despite this, the agents conducted lengthy vulnerability searches on their own and adapted their approach as the attacks unfolded.
By the end of last week, they had attempted to breach more than 1,200 websites. For the attacks that succeeded, the agents spent an average of about 23 hours continuously researching and testing the systems.
Fagerland estimates that the operation spent around $5,000 on AI tokens, or roughly $50 for each successful breach.
In his assessment, the hacker combined DeepSeek and Claude Code technologies.
Agents uncovered hidden vulnerabilities
One of the investigation’s most notable cases involved a large company whose public-facing application initially revealed no vulnerabilities to the agents.
Rather than stopping their search, they examined data and digital records to uncover parts of the infrastructure connected to the system. This led them to an internet-exposed control panel, where they identified a flaw in how data was processed.
After several hours of testing, the agents said they had exploited the vulnerability and forced the server to execute malicious commands.
“These kinds of solutions aren’t easy to find. This isn’t the work of amateur hackers,” Fagerland said.
Companies provide details about the incidents
Reuters was unable to determine whether the attacks had resulted in unauthorized data theft.
Deere said it had taken “necessary corrective measures” after the incident and that there was no evidence sensitive information had been accessed.
SharkNinja said a separate website containing only product support information had been affected, and that the threat had been neutralized.
Cummins said it was reviewing the report of a possible cyberattack against a website managed by one of its service providers.
Concern over autonomous hackers is growing
The use of autonomous AI agents to discover vulnerabilities, launch attacks and steal data is increasingly seen as a threat by the cybersecurity community.
In recent months, cases have been documented of cybercriminals using AI tools to breach organizations and obtain sensitive data.
Fagerland said he previously believed that hackers using AI agents would need at least some form of human oversight. This discovery, he said, shows that the technology has now crossed another threshold.
“Development in this field is extraordinary,” the cybersecurity researcher said.
