Similarly, we have Guarded for interactions with AI models; we should have something similar for RAG.

Examples of input guards for RAG:

• Encrypting text
• Homomorphing vectors
• Filtering per file types
• Verifying signature of files
• Sanitize files
• • Obfuscate sensitive data

Examples of output guards (before they are sent to LLM)

• Decrypt content
• Obfuscate sensitive data