Build the security layer for AI agents by modeling tools, data, permissions, and exposure into explainable risk signals and guardrails.