How Can I Research and Test AI Guardrails?

An Information Systems student is researching AI guardrails, aiming to implement and quantitatively compare protection methods against hate speech, misinformation, prompt injection, and data leakage. They considered LangChain and Google AI Studio, but Gemini's non-disableable guardrails pose a testing challenge. The student seeks alternatives for a more controlled testing environment without local LLM execution, and is open to suggestions on research structure, including a focused approach on data leakage using RAG.