Hacker Accessed More Than 100 Websites Using AI Agents

According to research by cybersecurity firm “Norforge,” a hacker accessed more than 100 websites using artificial intelligence agents that perform tasks largely autonomously. The targets included companies such as “John Deere,” “SharkNinja,” and “Cummins.” The hacker had very little involvement in the process, with AI agents doing most of the work. The incident shows that large-scale cyberattacks using artificial intelligence have already become a real threat.

Researchers discovered the attacks through work logs left publicly accessible on the “Cairn” platform. This open-source AI platform is described as a system that solves various problems. The user enters a starting point and an end goal, and the agents complete the task while recording every step. Hackers can use the same features to launch cyberattacks with multiple agents.

This individual was the most active attacker among the identified “Cairn” systems. The hacker targeted companies in the US, Canada, Australia, and Japan. The instruction given was fewer than 100 Chinese characters long, and the successful attacks took approximately 23 hours. “Norforge” found at least 21 publicly accessible “Cairn” systems suspected of being used for malicious purposes.

According to a “John Deere” representative, no evidence was found that sensitive data had been accessed during the attack. “Cummins” is investigating information about a possible cyberattack on a website managed by a third-party supplier. “SharkNinja” disclosed that only a separate website containing product support information was affected. The company also said it had contained the threat.

“Norforge” is working to shut down “Cairn” systems used for malicious purposes. The platform’s creator, known as “@le1xia0,” said that the “Cairn” system was never a hacking tool. According to the creator, misuse of the platform is not a feature of the system but a decision made by a person. The incident has renewed concerns about uncontrolled AI agents moving beyond testing environments and independently carrying out attacks.