Researchers developed LLM Honeypot, a novel system that monitors and identifies autonomous AI hacking agents targeting computer systems.

• Collected over 8 million hacking attempts during a three-month deployment
• Enhanced standard SSH honeypots with prompt injection techniques to detect LLM agents
• Employed time-based analysis to distinguish patterns specific to AI-powered attacks
• Identified and documented 8 potential LLM agent attacks in the wild

This research addresses the emerging security threat of LLM-powered attacks, providing security professionals with new detection methods and a deeper understanding of how AI agents operate in malicious contexts.

LLM Agent Honeypot: Monitoring AI Hacking Agents in the Wild