Definition
A technique where an AI agent retrieves relevant documents or data from a knowledge base before generating a response, grounding the model's output in factual sources. RAG reduces hallucination and enables the model to use proprietary or real-time data.
Why it matters
RAG is the dominant pattern for deploying large language models to enterprise data; however, RAG pipelines introduce new attack surfaces—poisoned documents in the knowledge base can inject false information or instructions into model responses.