1-2-3 Check: Enhancing Contextual Privacy in LLM via Multi-Agent Reasoning

TL;DR

This paper proposes a multi-agent framework to improve contextual privacy in large language models by decomposing privacy reasoning tasks, reducing leakage, and validating information flow, demonstrated on benchmark datasets.

Contribution

It introduces a novel multi-agent approach that decomposes privacy tasks, reducing information load and improving privacy adherence in LLMs, with systematic analysis and empirical validation.

Findings

Reduces private information leakage by 18-19% on benchmarks.

Outperforms single-agent baselines in privacy preservation.

Maintains public content fidelity while enhancing privacy.

Abstract

Addressing contextual privacy concerns remains challenging in interactive settings where large language models (LLMs) process information from multiple sources (e.g., summarizing meetings with private and public information). We introduce a multi-agent framework that decomposes privacy reasoning into specialized subtasks (extraction, classification), reducing the information load on any single agent while enabling iterative validation and more reliable adherence to contextual privacy norms. To understand how privacy errors emerge and propagate, we conduct a systematic ablation over information-flow topologies, revealing when and why upstream detection mistakes cascade into downstream leakage. Experiments on the ConfAIde and PrivacyLens benchmark with several open-source and closed-sourced LLMs demonstrate that our best multi-agent configuration substantially reduces private information leakage (\textbf{18\%} on ConfAIde and \textbf{19\%} on PrivacyLens with GPT-4o) while preserving the fidelity of public content, outperforming single-agent baselines. These results highlight the promise of principled information-flow design in multi-agent systems for contextual privacy with LLMs.