Isolation as a First-Class Principle for LLM-Agent System Safety: Concepts, Taxonomy, Challenges and Future Directions

PhD Qualifying Examination


Title: "Isolation as a First-Class Principle for LLM-Agent System Safety: 
Concepts, Taxonomy, Challenges and Future Directions"

by

Mr. Huihao JING


Abstract:

This survey treats isolation as a first-class principle for LLM-agent system 
safety. It argues that safety cannot be understood solely through 
input-output alignment because failures in agent systems often emerge from 
interactions across user inputs, tool access, execution channels, inter-agent 
communication, and environment- originated context. We organize the 
literature with a boundary- centric taxonomy of five boundaries: user-agent, 
agent-tool, agent-execution, agent-agent, and system-environment. This view 
helps explain where isolation is lost, how compromise spreads across 
boundaries, and which defenses are most relevant at each interface. We also 
summarize cross-boundary failure paths, discuss open challenges, and outline 
a research agenda for isolation-by- construction in future agent systems.


Date:                   Wednesday, 22 July 2026

Time:                   1:00pm - 2:00pm

Venue:                  Room 5501
                        Lift 25/26

Committee Members:      Dr. Yangqiu Song (Supervisor)
                        Dr. May Fung (Chairperson)
                        Dr. Xiaomin Ouyang