Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
October 20, 2025Open Access

Sanitize Your Responses: Mitigating Privacy Leakage in Large Language Models

View Full Paper
Ask AI
Bookmark
Share

Authors

WFWenjie FuHWHuandong WangJGJunyao Gao

Discussion

Loading...

Member takes

Overview

Novel framework reduces latency and computational overhead while improving privacy leakage control in LLMs.

Key Points

  • Self-Sanitize significantly enhances mitigation performance against privacy leakage in large language models.
  • The framework allows real-time streaming monitoring and repair, maintaining usability without substantial delays.
  • A lightweight Self-Monitor module inspects intentions at the token level, enabling effective privacy control.
  • Extensive experiments across four LLMs confirm the robustness of Self-Sanitize in diverse leakage scenarios.

Cite This Study

Fu et al. (2025) studied this question.

synapsesocial.com/papers/68f5fcce8d54a28a75cf1c1fhttps://doi.org/10.48550/arxiv.2509.24488
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1A Privacy-Preserving Middleware Architecture for Detecting Prompt Injection and Sensitive Data Exposure in Large-Language-Model Interactions2026
  2. 2A Framework for Real-time Safeguarding the Text Generation of Large Language Model2024
  3. 3Defend LLMs Through Self-Consciousness2025
  4. 4Developing Safe and Responsible Large Language Models -- A Comprehensive Framework2024 · 4 citations
  5. 5Robustifying Safety-Aligned Large Language Models through Clean Data Curation2024 · 1 citations