Skip to main content

Microsoft Copilot Studio Adds Human Approval for AI Agents

Microsoft is rolling out a new feature for Copilot Studio that puts a human in the loop for AI agents. Announced on September 3, this update lets developers require manual approval before an agent can carry out certain actions—like sending emails, closing support tickets, or handling financial transactions.

Why does this matter? As AI agents take on more responsibilities in business, the risk of them acting on their own—sometimes in ways that don't align with company policy—grows. Microsoft's answer is to add a checkpoint: when an agent wants to do something sensitive, it stops and asks for permission.

Here's how it works: the agent explains what it's about to do and why, and the user gets three choices—approve it, allow it just for this session, or reject it outright. This approval isn't just a suggestion; it's a hard rule that overrides the agent's own decision-making. Even if the AI thinks an action is the best next step, it has to wait for a human say-so.

To make this less disruptive, the approval requests show up right where people already work—inside Microsoft Teams and Microsoft 365 Copilot. No need to jump into the Copilot Studio backend to approve or deny. The feature is now generally available in the web version of Copilot Studio, delivered through Microsoft's standard multi-tenant cloud, and it's tracked under product roadmap number 570434.

This move reflects a broader trend in enterprise AI: finding the sweet spot between automation and oversight. Companies want the efficiency gains from AI, but they're also wary of the potential for mistakes or compliance slip-ups. By baking human approval into the workflow, Microsoft is offering a model that other AI platforms might follow.

For developers and IT teams, this means they can deploy AI agents with more confidence, knowing that critical actions won't happen without a human check. It's a practical step toward responsible AI deployment—one that acknowledges the limits of full autonomy.

Key Points

  • New manual approval feature in Microsoft Copilot Studio lets developers require human sign-off for AI agent actions.
  • Sensitive actions like sending emails, closing tickets, or processing payments trigger an approval request.
  • Users can approve, allow for the session, or reject the agent's proposed action.
  • Approval requests appear in Microsoft Teams and M365 Copilot, so users don't have to switch platforms.
  • Feature is generally available in the web version of Copilot Studio, via standard multi-tenant cloud.
  • Aims to balance automation efficiency with compliance and risk management in enterprise AI deployments.