Secure Multi-LLM Agentic AI and Zero-Trust Framework Survey

This document is a survey that discusses the integration of zero-trust security principles in multi-Large Language Model (multi-LLM) systems for Edge General Intelligence (EGI). It outlines the critical role of agentification in transforming edge devices into cognitive agents capable of complex task execution through collaboration. The survey identifies significant security vulnerabilities inherent in multi-LLM systems, such as insecure inter-LLM communications and potential data leakage. It presents a zero-trust framework that emphasizes the principle of 'never trust, always verify' to mitigate these vulnerabilities. The document categorizes zero-trust security mechanisms into model- and system-level approaches, detailing strategies such as strong identification and proactive maintenance. Additionally, it highlights the need for rigorous validation of data access requests and inter-LLM communications to prevent security breaches. The survey serves as a foundational treatment of zero-trust applications in multi-LLM contexts, offering both theoretical insights and practical strategies for enhancing security in EGI deployments.