Leveraging RAG for Training-Free Alignment of LLMs

Training-free LLM alignment using retrieval-augmented generation to improve refusal guardrails against agentic attacks without significant computational overhead.