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

2025-12-03

AI Summary

Existing privacy-preserving methods for educational multi-agent systems struggle to prevent PII leakage from LLM-mediated dialogues involving minors while preserving pedagogical semantic integrity. Method: We propose a Semantic Reconstruction Privacy Protection mechanism featuring a novel dual-stream architecture: a sanitization stream rigorously detects and de-identifies PII in unstructured dialogues, and a reconstruction stream leverages large language models to restore mathematical pedagogical logic and contextual coherence—decoupling privacy sanitization from semantic recovery. Contribution/Results: The mechanism guarantees zero sensitive-data leakage under a zero-trust paradigm while retaining instructional utility. Experiments on the MathDial dataset show that, when integrated with GPT-4o, it achieves 0% attack success rate and an exact-match accuracy of 0.8267—significantly outperforming pure-LLM baselines.

Technology Category

• Machine Learning: Privacy
• Natural Language Processing: Safety and Robustness
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems

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.

Problem

• Protecting minors' personal information in educational AI dialogues
• Balancing privacy security with teaching context preservation
• Preventing PII leakage in unstructured multi-agent learning systems

Innovation

• Dual-Stream Reconstruction Mechanism separates sanitization and context recovery
• LLM-driven context reconstruction preserves mathematical logic after PII removal
• Achieves zero PII leakage while maintaining high instructional accuracy