Microsoft is preparing to give makers a new safeguard for the Copilot Studio AI agents they build: the ability to require human approval before an agent is allowed to call a tool.
According to a report from Neowin, the update will let creators force their agents to pause and ask a person for sign-off before executing actions that reach outside their own environment. In practice, this means an agent could draft a response, prepare a document, or stage a transaction — but stop short of actually completing it until someone reviews and approves.
What’s changing for Copilot Studio AI agents
At its core, the feature adds a gate between an agent’s decision to act and the action itself. Many AI agents are built to be autonomous, pulling data, updating files, or triggering workflows without waiting for input. That autonomy is useful, but it also means an agent can take an irreversible or consequential step on its own.
The new approval requirement flips that model for the actions a maker chooses to protect. When enabled, the agent will call the tool but hold the result, surfacing it to a human who must explicitly approve before the action completes. This is the classic “human in the loop” pattern that enterprises have long relied on for anything risky — financial approvals, data changes, or communications sent on a company’s behalf.
Why “human in the loop” matters for agents
The push for human oversight reflects a broader shift across the AI industry. As agents gain the ability to take real-world actions rather than just answer questions, the line between “assistant” and “actor” blurs. A chatbot that only replies is low-risk; an agent that can submit a form, move money, or edit a shared file carries real consequences if it goes wrong.
By letting makers require approval on a per-action basis, Microsoft is giving organizations a way to keep agents useful while drawing a hard line at the steps that should never be fully automated. It also addresses a practical concern: when an agent does make a mistake, having a human review the action first makes it far easier to catch and correct before any damage occurs.
What this means for you
If you build or manage agents in Copilot Studio, this gives you a straightforward way to add governance without scrapping the automation you’ve already built. You can keep routine, low-risk steps running hands-off while routing anything consequential through a person. That’s especially relevant for agents that touch customer data, financial systems, or external communications.
For businesses still deciding whether to deploy agents broadly, the approval option lowers the barrier to trust. Instead of choosing between a fully autonomous agent and a heavily supervised one, you can mix both — letting agents do the heavy lifting and stepping in only where it counts.
How to get it
As of now, Microsoft has not announced a specific release date or build for the approval feature, and it isn’t available today. Makers will likely be able to enable it from the agent’s configuration settings once it rolls out, though the exact placement within Copilot Studio hasn’t been detailed.
For context, Copilot Studio — the low-code platform, formerly known as Power Virtual Agents, that sits within Microsoft’s Power Platform — has been Microsoft’s primary tool for letting organizations build custom copilots, chatbots, and agents without heavy coding. This approval capability fits naturally into that strategy, aiming to make agent deployment safer for enterprise use.
If you want to track its arrival, keep an eye on the official Microsoft Copilot documentation and the Microsoft 365 release notes, which typically list new Copilot Studio capabilities as they become available. Until then, this remains an announced-but-not-shipped feature.
Source: Neowin
Over to you: Would you trust your Copilot Studio agents to act fully autonomously, or do you want an approval step on every consequential action?
