This research introduces a robust framework for using LLMs with sensitive data by applying differential privacy to Retrieval-Augmented Generation (RAG) systems.

• Prevents sensitive information leakage when LLMs access protected databases
• Establishes theoretical privacy guarantees through novel DP-RAG algorithms
• Demonstrates practical implementation with minimal performance degradation
• Enables secure use of LLMs in healthcare, finance, and other privacy-critical domains

This innovation addresses a critical security challenge for enterprise AI adoption, allowing organizations to leverage powerful language models while maintaining strict data privacy compliance.

Privacy-Preserving Retrieval-Augmented Generation with Differential Privacy