A Chinese-speaking hacker has leveraged AI agents to hack into over 100 corporate websites worldwide over the past few weeks, according to local reports on October 19.

The affected companies include agricultural machinery maker John Deere, home appliance firm SharkNinja, and engine manufacturer Cummins.

In a report, Snorre Fagerland, chief researcher at cybersecurity firm Norfodh, explained that he discovered the hacker while scanning the web for instances where Cairn—an open-source AI problem-solving platform favored by hackers—was accidentally exposed to the internet.

Fagerland stated that the hacker instructed the AI agents to autonomously investigate and infiltrate target sites for several hours.

The instructions were given in Chinese using fewer than 100 characters.

Until last week, the AI agents attempted intrusions on more than 1,200 corporate websites, successfully breaching over 100 of them.

It took an average of 23 hours to achieve a successful breach.

Fagerland estimated that a mix of China's DeepSeek and Anthropic's Claude Code appeared to be used for the hacking, consuming a total of 5,000 dollars in token costs.

This translates to approximately 50 dollars spent per successful hack.

Fagerland stated that he has found 24 cases of Cairn being used for malicious activities since around October 2, noting that this allowed observers to watch hackers mobilize AI tools to launch attacks globally.

Affected company John Deere stated that it has taken necessary measures following the incident and that there are no signs the hacker accessed sensitive information.

SharkNinja also said the hack was limited to an independent site containing only product support information, and the threat has now been blocked.

Cummins stated that it is investigating a report regarding a potential cyberattack that occurred on a website operated by one of its external partners.

Fagerland emphasized that this discovery is a real-world example showing that commercial AI agents can carry out complex penetration processes from start to finish without human intervention, marking the opening of a completely new world.

It has not been confirmed how much data was leaked through this hacking.

The creator of Cairn stated in a recent post on X that Cairn is an innovative multi-agent architecture and was not designed for hacking attacks.

Fagerland explained that while he previously believed AI hacking agents would require minimal human supervision, it is now clear that the technology has moved past that stage, describing the advancement in this field as staggering.