Meet MemPrivacy: An Edge-Cloud Framework that Uses Local Reversible Pseudonymization to Protect User Data Without Breaking Memory Utility

Researchers introduced MemPrivacy, an edge-cloud framework for LLM agents that protects sensitive user data while keeping cloud memory useful.

The system replaces sensitive text on the user device with typed placeholders before cloud processing, then restores the original values locally afterward.

MemPrivacy also defines a four-level privacy taxonomy to decide which information needs protection and how strongly.

The approach aims to preserve personalized memory and response quality without sending raw private data to cloud systems.