Skip to main content

Microsoft reins in Copilot agents: human approval needed for emails and payments

Microsoft is putting a leash on its AI agents—at least for now. The tech giant announced that it will roll out updates to Copilot Studio in September, introducing a feature that requires human approval before AI agents can perform certain actions. Think of it as a safety brake for your digital assistant.

Copilot Studio is Microsoft's platform for building and managing AI agents, which can connect to various business tools like email, service ticketing systems, and payment gateways. While these agents are designed to streamline workflows, they can sometimes act too autonomously, which is great for routine tasks but risky when it comes to sensitive operations.

To address this, Microsoft is giving developers the option to set up a manual approval step for specific tools. With this new control, any action that requires human sign-off will be paused, and an approval request will pop up. The request will explain what the agent intends to do, and you can choose to approve it, approve it for the entire session, or reject it outright. It's like having a virtual assistant that asks before making a move—no more surprise emails or unauthorized payments.

Microsoft has highlighted three key scenarios where this approval mechanism will be particularly useful: sending emails, closing service tickets, and processing payments. Once these actions are configured with approval requirements, the AI agent won't be able to proceed on its own. This means you can breathe a little easier knowing that your AI won't accidentally send that embarrassing email to the entire company or approve a refund you didn't authorize.

The move comes amid growing concerns about AI autonomy and the potential for unintended consequences. By adding this human-in-the-loop feature, Microsoft is acknowledging that while AI can be incredibly helpful, it still needs oversight—especially when it comes to actions that have real-world impact.

For businesses, this feature offers a balance between automation and control. You can let your AI handle the mundane stuff, but when it comes to critical decisions, you're still in the driver's seat. It's a step towards more responsible AI deployment, and it might just give you more confidence in letting your digital agents take on more tasks.

So, if you're using Copilot Studio, keep an eye out for these updates in September. And remember, even the smartest AI sometimes needs a human to say, "Hold on, let me think about that."

Key Points

  • Microsoft will introduce manual approval requirements for Copilot Studio agents in September.
  • The feature applies to specific tools like sending emails, closing service tickets, and processing payments.
  • Users will see an approval request explaining the agent's intent, with options to approve, approve for the session, or reject.
  • This move aims to enhance safety and control, addressing concerns about AI autonomy in sensitive tasks.