Skip to content
AI News HubLIVE
Summary only

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

MemPrivacy is a framework from MemTensor, HONOR Device, and Tongji University that protects user privacy in LLM-powered agents by replacing sensitive data with typed placeholders on-device before cloud transmission, preserving memory utility. It introduces a four-level privacy taxonomy (PL1-PL4) with configurable masking thresholds. The best model achieves 85.97% F1 on MemPrivacy-Bench, and memory utility loss is limited to within 1.6% compared to up to 41.87% with irreversible masking.

SourceMarkTechPostAuthor: Asif Razzaq
Meet MemPrivacy: An Edge-Cloud Framework that Uses Local Reversible Pseudonymization to Protect User Data Without Breaking Memory Utility
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions

Summary only

Under the source content permission, this page provides an editorial summary, key points, and a link to the original article.

Key points and analysis

Article intelligence

EngineersIntermediate

Key points

  • MemPrivacy uses local reversible pseudonymization to replace sensitive data with typed placeholders before cloud processing, ensuring the cloud never sees raw private values.
  • A four-level privacy taxonomy (PL1-PL4) provides granular control over what gets protected, from general preferences to immediately exploitable credentials.
  • The MemPrivacy-4B-RL model achieves 85.97% F1 on MemPrivacy-Bench, outperforming general-purpose models like GPT-5.2 and Gemini-3.1-Pro.
  • Memory utility loss is kept within 1.6% at PL2-PL4 protection levels across three memory systems, compared to up to 41.87% loss with irreversible masking.

Highlights and analysis are generated automatically and may contain errors. Check the original source.

More recent stories on this topic · Models