Zenity

Zenity treats the agent as the unit of risk: who it acts for, what it can reach, and why it chose that action.

Coverage spans Copilot-class SaaS agents, homegrown agent platforms, and devices, with guardrails that fire before an over-permissioned agent ships. Labs research (AgentFlayer and Copilot Studio issues among them) feeds detections that prompt firewalls miss when the input looks harmless.