Commercially available AI agents have reportedly attempted to breach more than 1,200 websites. It raises concerns about their growing use in cybercrime. Researcher Fagerland said the tools could identify weaknesses and carry out complex attacks with little human involvement.
The campaign began on September 25 and targeted e-commerce websites in the US, Canada, Australia and Japan. Fagerland estimated that the operation cost about USD 5,000 in AI usage, or roughly USD 50 for each successful breach.
The Chinese-speaking hacker behind the campaign reportedly gave the AI agents brief instructions in fewer than 100 Chinese characters. By late last week, the agents had attempted to break into more than 1,200 websites.
Successful intrusions generally took around 23 hours of testing to identify security weaknesses. The campaign appeared to use a combination of China's DeepSeek and Anthropic's Claude Code, according to Fagerland's assessment.
In one case, the agents reportedly failed to find weaknesses in a company's public website. They then examined digital records, identified related infrastructure and located a publicly accessible control panel. After hours of testing, they reportedly exploited a weakness to make the server execute malicious commands.
‘It's not straightforward to come up with solutions like that,’ Fagerland said. ‘It's not script kiddie stuff.’ The researcher said the findings showed that commercially available AI tools could complete sophisticated intrusions without continuous human guidance.
Also Read: China’s Backdoor Warning Puts Anthropic’s Claude Code Under Global Security Spotlight
John Deere told Reuters that it had taken ‘necessary remediation actions’ after the incident. The company also said it had ‘no indication that sensitive information was accessed.’ SharkNinja said the attack affected a separate website that provides product support and confirmed 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.’ Reuters could not confirm how much data the attackers may have taken without permission, if any data was stolen during the incidents.
Fagerland said he had earlier believed that AI agents used for hacking would still need some human supervision. The latest findings have changed his view of what these tools can do on their own. ‘The development in this field is astounding,’ he said.
These incidents follow other reported cases of cybercriminals using AI tools to attack organizations. In August, Israeli cybersecurity firm Gambit Security found that Russian-speaking extortionists had used Cursor AI to break into seven organizations. In September, the firm reported another series of AI-assisted attacks that affected 25 companies. The attacks also led to the theft of at least 600,000 credit card records.