blog
AgentShield: A Deep Analysis of Deception-Based Runtime Intrusion Detection for Tool-Using LLM Agents
A deception-based compromise detection framework for tool-using LLM Agents. It transplants honeypot and honeytoken concepts from traditional network security into the Agent tool-call chain, achieving high detection rates with zero false positives through a three-layer progressive trap architecture.