Saurabh Deochake

ToolGuardian: Declarative Security for AI Agent-Tool Interactions

Arun Ravindran, Saurabh Deochake

LLM agents increasingly rely on external tools, expanding capability while creating a new security boundary: third-party tools may appear benign at the interface level while embedding unsafe behavior in implementation. Existing defenses rely on weak metadata, collapse characteriz…Read more