More Than 100 Websites Breached Using AI Agents

According to findings by security firm NorForge, a Chinese-speaking hacker utilized autonomous AI agents to infiltrate more than 100 websites, including those of companies in the US, Canada, Australia, and Japan. The attacks, which targeted well-known firms such as John Deere, SharkNinja, and Cummins, were determined to have been conducted via the open-source 'Cairn' AI problem-solving platform. Researchers noted that the campaign involved more than 1,200 infiltration attempts, with each successful breach taking an average of 23 hours.

Discovered by cybersecurity researcher Snorre Fagerland, the incident revealed that the hacker scanned exposed instances of the Cairn platform to coordinate the attacks. Estimated to have been carried out with a token cost of around $5,000, the operation reportedly used a mix of China's DeepSeek and Anthropic's Claude Code software. Successful breaches took approximately 23 hours, and the attacker was noted to have used fewer than 100 Chinese characters to execute the infiltration tactics.

John Deere, one of the targeted companies, stated that there was no evidence sensitive information was accessed and that necessary corrective steps had been taken. SharkNinja reported that the breach affected only an independent site containing product support information and that the threat had been contained. Meanwhile, Cummins announced that investigations are ongoing regarding a potential cyberattack on a website operated by a third-party vendor. @le1xia0, the creator of the Cairn platform, emphasized that the platform was never intended as a hacking tool and that the misuse was a human decision.