AgentSocialBench: Evaluating Privacy Risks in Human-Centered Agentic Social Networks
- URL: http://arxiv.org/abs/2604.01487v2
- Date: Mon, 06 Apr 2026 08:32:52 GMT
- Title: AgentSocialBench: Evaluating Privacy Risks in Human-Centered Agentic Social Networks
- Abstract summary: Privacy in agentic social networks is fundamentally harder than in single-agent settings.<n>cross-domain and cross-user coordination creates persistent leakage pressure.<n>Privacy instructions that teach agents how to abstract sensitive information paradoxically cause them to discuss it more.
- Score: 2.291770711277359
- License: http://creativecommons.org/licenses/by/4.0/
- Abstract: With the rise of personalized, persistent LLM agent frameworks such as OpenClaw, human-centered agentic social networks in which teams of collaborative AI agents serve individual users in a social network across multiple domains are becoming a reality. This setting creates novel privacy challenges: agents must coordinate across domain boundaries, mediate between humans, and interact with other users' agents, all while protecting sensitive personal information. While prior work has evaluated multi-agent coordination and privacy preservation, the dynamics and privacy risks of human-centered agentic social networks remain unexplored. To this end, we introduce AgentSocialBench, the first benchmark to systematically evaluate privacy risk in this setting, comprising scenarios across seven categories spanning dyadic and multi-party interactions, grounded in realistic user profiles with hierarchical sensitivity labels and directed social graphs. Our experiments reveal that privacy in agentic social networks is fundamentally harder than in single-agent settings: (1) cross-domain and cross-user coordination creates persistent leakage pressure even when agents are explicitly instructed to protect information, (2) privacy instructions that teach agents how to abstract sensitive information paradoxically cause them to discuss it more (we call it abstraction paradox). These findings underscore that current LLM agents lack robust mechanisms for privacy preservation in human-centered agentic social networks, and that new approaches beyond prompt engineering are needed to make agent-mediated social coordination safe for real-world deployment.
Related papers
- Synergy: A Next-Generation General-Purpose Agent for Open Agentic Web [23.792766952415317]
We argue that the next generation of AI agents must become Agentic Citizens.<n>We present Synergy, a general-purpose agent architecture and runtime harness for persistent, collaborative, and evolving agents.
arXiv Detail & Related papers (2026-03-30T13:35:37Z) - OMNI-LEAK: Orchestrator Multi-Agent Network Induced Data Leakage [59.3826294523924]
We investigate the security vulnerabilities of a popular multi-agent pattern known as the orchestrator setup.<n>We report the susceptibility of frontier models to different categories of attacks, finding that both reasoning and non-reasoning models are vulnerable.
arXiv Detail & Related papers (2026-02-13T21:32:32Z) - The Rise of AI Agent Communities: Large-Scale Analysis of Discourse and Interaction on Moltbook [62.2627874717318]
Moltbook is a Reddit-like social platform where AI agents create posts and interact with other agents through comments and replies.<n>Using a public API snapshot collected about five days after launch, we address three research questions: what AI agents discuss, how they post, and how they interact.<n>We show that agents' writing is predominantly neutral, with positivity appearing in community engagement and assistance-oriented content.
arXiv Detail & Related papers (2026-02-13T05:28:31Z) - AgentCrypt: Advancing Privacy and (Secure) Computation in AI Agent Collaboration [23.22266919684932]
AgentCrypt is a framework for fine-grained, encrypted agent communication.<n>It ensures privacy across diverse interactions and enables computation on otherwise inaccessible data.<n>We implemented and tested it with Langgraph and Google ADK, demonstrating versatility across platforms.
arXiv Detail & Related papers (2025-12-08T23:20:20Z) - ConVerse: Benchmarking Contextual Safety in Agent-to-Agent Conversations [11.177126931962443]
ConVerse is a benchmark for evaluating privacy and security risks in agent-agent interactions.<n>It spans three practical domains with 12 user personas and over 864 contextually grounded attacks.<n>By unifying privacy and security within interactive multi-agent contexts, ConVerse reframes safety as an emergent property of communication.
arXiv Detail & Related papers (2025-11-07T15:49:49Z) - Multi-Hop Privacy Propagation for Differentially Private Federated Learning in Social Networks [1.5653612447564105]
Federated learning enables collaborative model training across decentralized clients without sharing local data.<n>A client's privacy loss depends not only on its privacy protection strategy but also on the privacy decisions of others.<n>We propose a socially-aware privacy-preserving FL mechanism that systematically quantifies indirect privacy leakage.
arXiv Detail & Related papers (2025-08-11T06:53:32Z) - Agentic Web: Weaving the Next Web with AI Agents [109.13815627467514]
The emergence of AI agents powered by large language models (LLMs) marks a pivotal shift toward the Agentic Web.<n>In this paradigm, agents interact directly with one another to plan, coordinate, and execute complex tasks on behalf of users.<n>We present a structured framework for understanding and building the Agentic Web.
arXiv Detail & Related papers (2025-07-28T17:58:12Z) - AI Delegates with a Dual Focus: Ensuring Privacy and Strategic Self-Disclosure [40.47039082007319]
Large language model (LLM)-based AI delegates are increasingly utilized to act on behalf of users.<n>We propose a novel AI delegate system that enables privacy-conscious self-disclosure.<n>Our user study demonstrates that the proposed AI delegate strategically protects privacy, pioneering its use in diverse and dynamic social interactions.
arXiv Detail & Related papers (2024-09-26T08:45:15Z) - Knowledge Boundary and Persona Dynamic Shape A Better Social Media Agent [69.12885360755408]
We construct a social media agent based on personalized knowledge and dynamic persona information.
For personalized knowledge, we add external knowledge sources and match them with the persona information of agents, thereby giving the agent personalized world knowledge.
For dynamic persona information, we use current action information to internally retrieve the persona information of the agent, thereby reducing the interference of diverse persona information on the current action.
arXiv Detail & Related papers (2024-03-28T10:01:23Z) - Cross-Network Social User Embedding with Hybrid Differential Privacy
Guarantees [81.6471440778355]
We propose a Cross-network Social User Embedding framework, namely DP-CroSUE, to learn the comprehensive representations of users in a privacy-preserving way.
In particular, for each heterogeneous social network, we first introduce a hybrid differential privacy notion to capture the variation of privacy expectations for heterogeneous data types.
To further enhance user embeddings, a novel cross-network GCN embedding model is designed to transfer knowledge across networks through those aligned users.
arXiv Detail & Related papers (2022-09-04T06:22:37Z) - Aligning to Social Norms and Values in Interactive Narratives [89.82264844526333]
We focus on creating agents that act in alignment with socially beneficial norms and values in interactive narratives or text-based games.
We introduce the GALAD agent that uses the social commonsense knowledge present in specially trained language models to contextually restrict its action space to only those actions that are aligned with socially beneficial values.
arXiv Detail & Related papers (2022-05-04T09:54:33Z)
This list is automatically generated from the titles and abstracts of the papers in this site.
This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.