Paper 2605.09530
MemPrivacy: Privacy-Preserving Personalized Memory Management for Edge-Cloud Agents
- Published
- May 2026
- Research lab
- Independent
- Citations
- 7
- GitHub
- 120 stars
01 In brief
Summary
MemPrivacy is a framework for privacy-preserving memory management in edge-cloud LLM agents.
It detects privacy-sensitive spans on edge devices, replaces them with type-aware placeholders (e.g., <Health_Info_1>) before cloud processing, and restores original values locally.
This decouples privacy protection from semantic destruction, preserving memory utility.
The framework introduces a four-level privacy taxonomy (PL1–PL4) for configurable protection and MemPrivacy-Bench, a benchmark with 200 users and over 155k privacy instances.
Lightweight models (0.6B–4B) are trained via SFT and GRPO.
Experiments show MemPrivacy outperforms general LLMs (e.g., GPT-5.2, Gemini-3.1-Pro) in privacy extraction, with F1 scores up to 85.97% on MemPrivacy-Bench and 94.48% on PersonaMem-v2, while reducing latency.
Across memory systems (LangMem, Mem0, Memobase), utility loss is limited to within 1.6%, outperforming masking baselines.
MemPrivacy balances privacy and utility for edge-cloud agents, enabling secure, practical deployment.
02 From the paper
Abstract
As LLM-powered agents are increasingly deployed in edge-cloud environments, personalized memory has become a key enabler of long-term adaptation and user-centric interaction. However, cloud-assisted memory management exposes sensitive user information, while existing privacy protection methods typically rely on aggressive masking that removes task-relevant semantics and consequently degrades memory utility and personalization quality. To address this challenge, We propose MemPrivacy, which identifies privacy-sensitive spans on edge devices, replaces them with semantically structured type-aware placeholders for cloud-side memory processing, and restores the original values locally when needed. By decoupling privacy protection from semantic destruction, MemPrivacy minimizes sensitive data exposure while retaining the information required for effective memory formation and retrieval. We also construct MemPrivacy-Bench for systematic evaluation, a dataset covering 200 users and over 155k privacy instances, and introduce a four-level privacy taxonomy for configurable protection policies. Experiments show that MemPrivacy achieves strong performance in privacy information extraction, substantially surpassing strong general-purpose models such as GPT-5.2 and Gemini-3.1-Pro, while also reducing inference latency. Across multiple widely used memory systems, MemPrivacy limits utility loss to within 1.6%, outperforming baseline masking strategies. Overall, MemPrivacy offers an effective balance between privacy protection and personalized memory utility for edge-cloud agents, enabling secure, practical, and user-transparent deployment.