Retrieval-Augmented Generation (RAG) enhances LLMs by providing external knowledge, but introduces new security and trust challenges.

• Reduces hallucinations by grounding LLM outputs in retrieved context
• Improves factuality through integration of up-to-date information
• Introduces new vulnerabilities to adversarial attacks and robustness issues
• Requires security frameworks to ensure safe deployment in critical applications

This research establishes a comprehensive trustworthiness framework for RAG systems, essential for deploying secure AI solutions in enterprise environments where information accuracy and system resilience are paramount.

Towards Trustworthy Retrieval Augmented Generation for Large Language Models: A Survey