Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SRPG: Semantically Reconstructed Privacy Guard for Zero-Trust Privacy in Educational Multi-Agent Systems

Shuang Guo, Zihui Li

TL;DR

SRPG tackles privacy in educational multi-agent systems by decoupling private data from instructional content through a dual-stream reconstruction mechanism. It combines a Strict Sanitization Stream that guarantees zero PII leakage with a Context Reconstruction Stream that recovers essential mathematical context, enabling effective tutoring. Empirical evaluation on MathDial shows zero Attack Success Rate across backbones and strong math-utility, with GPT-4o achieving 0.0000 ASR and 0.8267 Exact Match, surpassing baselines. The approach demonstrates a practical, future-proof method for protecting minors' privacy without sacrificing educational quality, with potential extensions to other domains.

Abstract

Multi-Agent Systems (MAS) with large language models (LLMs) enable personalized education but risk leaking minors personally identifiable information (PII) via unstructured dialogue. Existing privacy methods struggle to balance security and utility: role-based access control fails on unstructured text, while naive masking destroys pedagogical context. We propose SRPG, a privacy guard for educational MAS, using a Dual-Stream Reconstruction Mechanism: a strict sanitization stream ensures zero PII leakage, and a context reconstruction stream (LLM driven) recovers mathematical logic. This decouples instructional content from private data, preserving teaching efficacy. Tests on MathDial show SRPG works across models; with GPT-4o, it achieves 0.0000 Attack Success Rate (ASR) (zero leakage) and 0.8267 Exact Match, far outperforming the zero trust Pure LLM baseline (0.2138). SRPG effectively protects minors privacy without sacrificing mathematical instructional quality.

SRPG: Semantically Reconstructed Privacy Guard for Zero-Trust Privacy in Educational Multi-Agent Systems

TL;DR

SRPG tackles privacy in educational multi-agent systems by decoupling private data from instructional content through a dual-stream reconstruction mechanism. It combines a Strict Sanitization Stream that guarantees zero PII leakage with a Context Reconstruction Stream that recovers essential mathematical context, enabling effective tutoring. Empirical evaluation on MathDial shows zero Attack Success Rate across backbones and strong math-utility, with GPT-4o achieving 0.0000 ASR and 0.8267 Exact Match, surpassing baselines. The approach demonstrates a practical, future-proof method for protecting minors' privacy without sacrificing educational quality, with potential extensions to other domains.

Abstract

Multi-Agent Systems (MAS) with large language models (LLMs) enable personalized education but risk leaking minors personally identifiable information (PII) via unstructured dialogue. Existing privacy methods struggle to balance security and utility: role-based access control fails on unstructured text, while naive masking destroys pedagogical context. We propose SRPG, a privacy guard for educational MAS, using a Dual-Stream Reconstruction Mechanism: a strict sanitization stream ensures zero PII leakage, and a context reconstruction stream (LLM driven) recovers mathematical logic. This decouples instructional content from private data, preserving teaching efficacy. Tests on MathDial show SRPG works across models; with GPT-4o, it achieves 0.0000 Attack Success Rate (ASR) (zero leakage) and 0.8267 Exact Match, far outperforming the zero trust Pure LLM baseline (0.2138). SRPG effectively protects minors privacy without sacrificing mathematical instructional quality.

Paper Structure

This paper contains 18 sections, 4 equations, 2 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLM-based Multi-Agent Systems
  4. Text Privacy Preservation and Reconstruction
  5. Federated and Embedded Privacy
  6. Preliminary
  7. Methodology
  8. Problem Formulation
  9. The SRPG Framework
  10. Strict Sanitization Stream
  11. Context Reconstruction Stream
  12. Dual-Stream Fusion
  13. Experiments
  14. Experimental Setup
  15. Main Results
  16. ...and 3 more sections

Figures (2)

  • Figure 1: Performance Comparison of Privacy-Preserving Methods (SRPG vs. Baselines). SRPG achieves absolute privacy (ASR 0.00) while maintaining superior utility compared to Naive Masking.
  • Figure 2: Performance of SRPG under Different Backbone Models (Ablation Study). Note the significant jump in Exact Match score with GPT-4o.