The "Foot-In-The-Door" (FITD) jailbreak represents a sophisticated multi-turn prompt injection attack targeting Large Language Models (LLMs). This technique exploits the psychological principle of incremental commitment, gradually escalating malicious requests to circumvent established safety mechanisms within LLMs. By utilizing intermediate "bridge" prompts and self-alignment strategies, the attack methodically persuades the model to produce increasingly harmful outputs, even when the model initially declines similar direct requests. Mitigation steps: **For AI Developers:** * Enhance user input validation and filtering to detect and mitigate malicious intent in multi-turn dialogue, including techniques that specifically address the "foot-in-the-door" pattern. * Implement robust multi-turn conversation safety mechanisms to detect and prevent gradual escalation of malicious intent. * Incorporate sophisticated detection mechanisms that account for paraphrasing and contextual shifting of requests. **For Model Trainers/Fine-tuners:** * Develop improved alignment techniques resilient to iterative manipulation and "self-corruption" of model outputs. * Regularly audit LLMs for vulnerabilities using advanced automated red-teaming techniques that simulate multi-turn interactions.