HoneyTrap – A New LLM Defense Framework to Counter Jailbreak Attacks

This report introduces HoneyTrap, a multi-agent deceptive defense framework to counter multi-turn jailbreak attacks against large language models, combining a Threat Interceptor, Misdirection Controller, System Harmonizer, and Forensic Tracker to dynamically mislead attackers while preserving benign user experience. Evaluated on GPT-4, GPT-3.5-turbo, Gemini-1.5-pro, and LLaMa-3.1, it reportedly cuts attack success rates by an average of 68.77%, boosts mislead success by ~118%, and increases attacker resource consumption by 149% without degrading normal response quality.