Chinese Hacker Used AI Agents to Attack More Than 1,200 Websites

Researchers discovered that a hacker acted alone to use AI agents for mass attacks on websites. More than 1,200 sites were targeted, and successful compromises affected resources linked to over 100 companies, including John Deere, SharkNinja and Cummins.

Estimated spending on AI tokens was about $5,000, or roughly $50 per successful compromise. Successful attacks took about 23 hours on average, while some tasks ran for more than five days.

The tool used was the open-source multi-agent framework Cairn. Based on worker names, researchers inferred that the underlying stack likely combined Claude Code and DeepSeek V4.1 Flash. These tools are usually used for coding, but here they were repurposed for large-scale attacks.

To operate Cairn, a human only needed to specify a target and provide instructions. The AI agents then independently searched for footholds, attempted exploits, recorded results and decided what to do next.

Researchers were able to uncover the hidden activity because the attacker left a control panel publicly accessible on the internet. Through these panels, they could study the agents’ tasks and execution logs.

The irony is clear: the hacker tasked AI with finding vulnerabilities in other systems, but failed to secure his own control infrastructure. The case shows that autonomous AI agents can already become a practical tool for mass cyberattacks.