| name | pisas-contextual-integrity-multi-user-agentic |
| description | Benchmarking contextual integrity in multi-user agentic systems. As LLM agents evolve into shared organizational infrastructure, new privacy risks emerge from inter-agent messages, shared memory, and cross-user information exposure. Activation: contextual integrity, multi-user agents, privacy benchmark, agentic privacy, inter-agent communication, shared memory privacy. |
| metadata | {"arxiv_id":"2607.05318","published":"2026-07-06","authors":"Shubham Gupta, Nazanin Mohammadi Sepahvand, Abhinav Kumar, Cem Subakan, Spandana Gella, Pierre-André Noël, Perouz Taslakian, Eugene Bagdasarian, Valentina Zantedeschi","tags":["contextual-integrity","multi-user-agents","privacy","benchmark","inter-agent-communication","shared-memory"]} |
PiSAs: Benchmarking Contextual Integrity in Multi-User Agentic Systems
Overview
As LLM agents evolve from single-user assistants into shared organizational infrastructure, new privacy risks emerge: inappropriate information may not only be exposed through outputs for external recipients, but also internally across users through inter-agent messages, shared memory and agents. This paper introduces PiSAs, a benchmark for evaluating contextual integrity in multi-user agentic systems.
Key Problem
Multi-User Privacy Risks in Agentic Systems
- Inter-agent message leakage: Agents sharing information with other agents serving different users
- Shared memory exposure: Common memory stores accessible across user contexts
- Contextual integrity violations: Information appropriate in one context may be inappropriate in another
- Internal vs external risks: Privacy violations can occur internally (across users) not just externally
Contextual Integrity Framework
- Based on contextual integrity theory: information flow norms depend on context, sender, recipient, and information type
- Agents must respect contextual boundaries when processing and sharing information
- Multi-user settings introduce complex information flow constraints
Benchmark Design
- Multi-User Scenarios: Agents serving different users with overlapping but distinct information access
- Privacy Violation Detection: Measures for detecting contextual integrity violations
- Inter-Agent Communication: Tests for information leakage through agent-to-agent messages
- Shared Memory Access: Evaluates privacy preservation in shared memory architectures
Implications
- New privacy paradigm needed for multi-user agentic AI systems
- Contextual integrity as a framework for designing privacy-preserving agent architectures
- Benchmark enables systematic evaluation of privacy risks in deployed multi-user agents
- Organizational deployments of LLM agents need context-aware information flow controls
Pitfalls
- Contextual integrity norms are domain-specific and may not generalize
- Benchmark scenarios may not capture all real-world privacy failure modes
- Trade-off between privacy and agent utility/efficiency
- Dynamic context boundaries are hard to define in practice
Activation Keywords
contextual integrity, multi-user agents, privacy benchmark, agentic privacy, inter-agent communication, shared memory privacy, organizational agents, information flow norms, PiSAs
Paper Reference
arXiv:2607.05318 - "PiSAs: Benchmarking Contextual Integrity in Multi-User Agentic Systems" (Jul 2026)