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

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 study over information-flow topologies, revealing when and why upstream detection mistakes cascade into downstream leakage. The 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 18% on ConfAIde and 19% on PrivacyLens using GPT-4o) while preserving the fidelity of public content, outperforming single-agent baselines. These results highlight the promise of principled, visibility-aware information-flow design in multi-agent systems for contextual privacy with LLMs.