A Chinese-speaking hacker has been observed using artificial intelligence agents to break into more than 100 websites belonging to well-known companies, according to a report from Reuters. The attacks leveraged exposed instances of Carin, an open-source AI problem-solving platform, to probe and exploit vulnerabilities in targets including farming equipment manufacturer Deere and kitchen appliance company SharkNinja.
Security researcher Snorre Fagerland, principal threat researcher at cybersecurity firm NorForge, spotted the hacker's activities while monitoring Carin instances that had been inadvertently exposed. Fagerland observed multiple hackers at work, but identified the Chinese-speaking hacker as the most prolific.
The hacker targeted businesses based in the United States, Canada, Australia, and Japanese e-commerce sites. Remarkably, the commands given to the AI agents used fewer than 100 Chinese characters, yet the agents had attempted to break into more than 1,200 different websites as of the week ending October 18, 2026.
Fagerland expressed surprise at the sophistication of some attacks. In one case, AI agents scoured digital records to find infrastructure associated with their target, located a public-facing control panel, identified a weakness, and then leveraged it to force the execution of malicious commands.
"It's not straightforward to come up with solutions like that," Fagerland told Reuters. "It's not script kiddie stuff."
The approach using AI agents means attackers can issue instructions and leave the agents to operate without supervision.
SharkNinja told Reuters that the attack only impacted a site used for product support information. However, other AI-aided hackers monitored by Fagerland had more success. One stole credit card data, while another attempted to steal from a cryptocurrency exchange, though it remains unclear whether that attempt succeeded.
Analysis
Why This Matters
- AI agents are enabling a new class of automated cyberattacks that can operate without direct human supervision
- The low cost of entry (fewer than 100 characters of instructions) lowers the skill barrier for sophisticated hacking
- Companies may need to reassess exposure of AI tools and public-facing infrastructure as attackers weaponize the same platforms
Background
This incident represents an escalation in the use of generative AI for malicious purposes. While hackers have long used automation for reconnaissance and brute-force attacks, the ability of AI agents to independently identify vulnerabilities, navigate systems, and execute multi-step exploits marks a shift. Carin, the open-source platform observed in these attacks, has become popular among hackers for its problem-solving capabilities. The attacks mirror a broader trend of cybersecurity researchers warning that offensive AI capabilities are advancing faster than defensive measures.
Key Perspectives
Security Researchers: Snorre Fagerland of NorForge observed the attacks directly and noted the sophistication of the AI agents' tactics. The researcher was "impressed" by the agents' ability to chain together multiple steps, from infrastructure discovery to exploit execution, without human guidance.
Affected Companies: SharkNinja confirmed the attack but stated it only affected a site used for product support information, suggesting limited damage. However, other targets may have suffered more serious breaches, including theft of credit card data and attempted cryptocurrency theft.
Critics and Skeptics: While the attack demonstrates capability, the long-term risk depends on whether defensive tools can keep pace. The use of exposed AI instances also highlights that organizations may inadvertently provide the tools attackers need, raising questions about secure deployment of open-source AI platforms.
What to Watch
- Whether more companies report breaches linked to similar AI-driven attacks in coming weeks
- How platform operators of open-source AI tools like Carin respond to prevent abuse of exposed instances
- Potential regulatory or law enforcement response given the multi-country scope of the targeting
