PAAC: Privacy-Aware Agentic Device-Cloud Collaboration

2026-05-08

AI Summary

This work addresses the challenge of reconciling strong reasoning capabilities with user privacy in device–cloud collaborative large language model (LLM) agents. The authors propose treating the device–cloud boundary as a trust boundary and innovatively aligning a planner–executor architecture with this system partition: the cloud processes de-identified tasks containing typed placeholders, while the device identifies sensitive information and generates executable summaries. By integrating sensitive fragment detection, typed placeholder substitution, and a deterministic registry mechanism, the approach preserves the structural integrity of tool invocations while enabling flexible, executable privacy protection. Evaluated on three stringent privacy benchmarks, the method dominates the Pareto frontier, achieving 15–36% higher average accuracy and 2–6× lower information leakage, and demonstrates consistently superior performance across 17 diverse cross-domain tasks.

Abstract

Large language model (LLM) agents face a structural tension: cloud agents provide strong reasoning but expose user data, while on-device agents preserve privacy at the cost of overall capability. Existing device-cloud designs treat this boundary as a compute split rather than a trust boundary suited to agentic workloads, and existing sanitizers force a choice between policy flexibility and the structural fidelity tool calls require. In this work, we develop PAAC, a privacy-aware agentic framework that aligns planner--executor decomposition with the device-cloud boundary so that role specialization itself becomes the privacy mechanism. The cloud agent reasons over typed placeholder tokens that preserve each sensitive value's reasoning role while discarding its content, while the on-device agent identifies sensitive spans and distills each step's execution outcome into compact key findings. Sanitization confines the on-device LLM to proposing which spans to mask, while a deterministic registry performs all substitution and reversal, keeping actions directly executable on device. On three agentic benchmarks under strict privacy settings, PAAC dominates the Pareto frontier of privacy and accuracy, improving average accuracy by 15-36\% and reducing average leakage by 2-6$\times$ over state-of-the-art device-cloud baselines, with the largest margins on privacy targets outside fixed entity taxonomies. We find consistent improvements on 17 additional benchmarks spanning 10 domains, including math, science, and finance.

Problem

• privacy-aware
• device-cloud collaboration
• LLM agents
• trust boundary
• data sanitization

Innovation

• privacy-aware collaboration
• device-cloud LLM agents
• typed placeholder tokens
• planner-executor decomposition
• deterministic sanitization registry