Survey frames isolation as core safety principle for LLM-agent systems, arXiv preprint finds

A survey preprint on arXiv proposes treating isolation as a first-class safety principle for large language model agent systems, arguing that current safety literature is fragmented across attack types and applications. The authors define isolation as the separation of user inputs, tool access, execution channels, inter-agent communication, and environment-originated context. They organize existing work into a boundary-centric taxonomy spanning five interfaces: user-agent, agent-tool, agent-execution, agent-agent, and system-environment. The framework aims to identify where isolation loss first occurs, how compromise propagates across boundaries, and which defenses are most relevant at each interface. The paper also summarizes cross-boundary failure paths, discusses open challenges, and outlines a research agenda for isolation-by-construction in future agent systems.