AgentGuard: Building a Trust Boundary for Agentic Execution
AI agents are becoming increasingly capable of taking actions rather than simply generating text.
They can inspect systems, call tools, execute code, modify state, recover from failures, and interact
