Microsoft Adds Human Approval Controls for Sensitive Actions in Copilot Studio
Microsoft is updating Copilot Studio to allow developers to require human approval before AI agents execute specific tools or actions. When a sensitive action—such as sending emails, closing support tickets, or processing payments—is triggered, the agent pauses and displays its intent for user approval or rejection. This human-in-the-loop mechanism enhances AI safety and enterprise governance by preventing autonomous agents from accidentally executing high-risk or irreversible business operations. It helps organizations adopt AI-driven autonomous workflows with greater confidence and operational control. Developers can configure approval settings on a per-agent and per-tool basis using dedicated toggles in Copilot Studio. When an approval prompt appears, users can choose to approve the single execution, approve the action for the entire session, or reject the request entirely.
## BACKGROUND
Microsoft Copilot Studio is a low-code platform that enables enterprise developers to build, test, and manage conversational and autonomous AI agents. Modern AI agents are capable of connecting to enterprise systems like email servers, ticketing platforms, and payment gateways to execute multi-step tasks independently. However, fully autonomous execution carries risks if an agent misinterprets instructions, making human supervision essential for high-stakes actions.