Foot In The Door: Understanding Large Language Model Jailbreaking via Cognitive Psychology