Pac Bench Privacy Multi Agent

---
name: pac-bench-privacy-multi-agent
description: 'Research paper: PAC-BENCH - Evaluating Multi-Agent Collaboration under Privacy Constraints. First benchmark for differential privacy in multi-agent systems.'
metadata:
  openclaw:
    emoji: "🔐"
    tags: ["research", "arxiv", "privacy", "multi-agent", "benchmark", "coordination", "differential-privacy"]
---

# PAC-BENCH: Evaluating Multi-Agent Collaboration under Privacy Constraints

**arXiv ID:** 2604.15871  
**Published:** 2026-04-17  
**Authors:** Minjun Park, Donghyun Kim, Hyeonjong Ju  
**Categories:** cs.AI, cs.MA  
**Utility Score:** 0.95

## Abstract

We are entering an era in which individuals and organizations increasingly deploy dedicated AI agents that interact and collaborate with other agents. However, the dynamics of multi-agent collaboration under privacy constraints remain poorly understood. We introduce PAC-BENCH, the first benchmark for evaluating multi-agent collaboration under differential privacy constraints.

## Key Contributions

1. **Privacy-Preserving Collaboration**: First benchmark for multi-agent coordination with differential privacy
2. **Cross-Organizational Agents**: Evaluates agents from different organizations collaborating privately
3. **Privacy-Utility Tradeoff**: Measures the impact of privacy constraints on collaboration effectiveness

## Relevance to AI Agent Systems

- **Multi-Agent Coordination**: Privacy-preserving agent collaboration
- **AI Agents**: Security and privacy for deployed agents
- **Evaluation**: Benchmarking privacy-utility tradeoffs
- **Coordination**: Cross-organizational agent workflows

## Technical Keywords

multi-agent, ai agent, ai agents, benchmark, evaluation, coordination

## URL

http://arxiv.org/abs/2604.15871

---

**Tracked:** 2026-04-20  
**Source:** arXiv Paper Tracker