Building a Hybrid Rule-Based and Machine Learning Framework to Detect and Defend Against Jailbreak Prompts in LLM Systems