Deelen iwwer


Human supervision of computer use (preview)

[This article is prerelease documentation and is subject to change.]

When you use human supervision, the computer-using agent escalates to the person you configure in the Human supervision settings when it needs confirmation or more information.

Note

Computer-use agents might encounter prompt injection attacks, where hidden instructions in screenshots, web pages, or other inputs attempt to influence actions in unintended ways. To minimize this risk, operate these agents within trusted, isolated environments and apply robust validation checks before executing any instructions.

Important

This article contains Microsoft Copilot Studio preview documentation and is subject to change.

Preview features aren't meant for production use and may have restricted functionality. These features are available before an official release so that you can get early access and provide feedback.

If you're building a production-ready agent, see Microsoft Copilot Studio Overview.

Feature support by model

The following table lists which models support the human supervision feature.

Model provider Version Status Supports human supervision
Azure OpenAI OpenAI Computer Using Agent V1 Retired Yes
Azure OpenAI Computer-Using Agent (CUA) Active Yes
Anthropic Claude Sonnet 4.5 Active Yes

Review agent and computer-use activity

During execution, when the computer-using agent needs confirmation or additional information, it may send a review request to the configured human reviewer via Outlook email and displays an inline review card in the Advanced computer-use activity panel.

Human review requests are triggered by probabilistic AI model behavior, which means they may not trigger in every situation where a person would want a pause, and they may also trigger when a pause isn't necessary. You should not rely on human review or clarification requests as a safety fail-safe or as a guarantee that the system will always request human input before proceeding.

Note

  • Only the maker of the computer use tool receives human review requests and has access to verbose logs through the Advanced computer-use activity panel. If you have set End user credentials under machine settings, these requests are sent to the person who originally set up the connection, but only if they're also included as one of the designated Human supervisors.
  • To review the human review requests inline in the computer-using agent activity side panel within the agent's activity map, you need to enable the Advanced computer-use activity feature in your environment. If the advanced computer-use activity experience doesn't appear in the activity map, your administrator might have turned off this feature for your environment. Learn more in Advanced computer-use activity.

The following table shows the main information included in a human review email request:

Field Description
Agent using the tool The name of the Copilot Studio agent.
Computer-use tool The name of the computer-use tool that triggered the request.
Conversation The deep-link to the agent activity where the computer use tool is currently running in.
Request timeout The request timeout date and time
Requested by The maker (connection owner) of the computer-use tool.
Provide missing information The textbox for the response to submit to the computer-use tool.
Submit The button to submit the entered text response. In case of confirmation requests, either type Yes to confirm or No to decline.

Respond to human review requests

To respond to human review requests, you can either respond by email or inline using the computer‑use activity panel in the agent's activity map. Inline responses require the Advanced computer‑use activity panel to be enabled in the environment. In both cases, if you don't provide a response, the workflow remains paused and automatically stops when the timeout specified in the email is reached.

Important

The human supervision system is based on probabilistic AI model behavior. While the system includes a system prompt and perhaps other constraints, it might occasionally generate clarification requests that are unnecessary. It might fail to trigger when one is expected, or it might ask for information that shouldn't be provided in the given context, including potentially sensitive information.

As a result, reviewers should apply judgment when responding to human supervision requests and shouldn't submit sensitive information such as usernames, passwords, PINs, credit card information, or Social Security numbers. Depending on your organization's configuration, responses to human review requests might be persisted in your Dataverse environment and/or Purview audit logs and might be visible to authorized users within your organization.

Respond via email

The human review request email includes details about the agent, the tool, and the conversation, along with the specific information or clarification being requested. Review the request carefully to determine whether you have sufficient information to respond. If appropriate, enter your response in the designated text field and select Submit.

Respond inline

  1. Select the Conversation link in the human review request email to open the agent's activity map.
  2. Find the computer‑use action that matches the configured computer‑use tool, and open its side panel.
  3. In the Activity section of the side panel, review the list of activities. Activities awaiting human review are marked with a waiting indicator next to the model response.
  4. Review the request carefully and determine whether you have sufficient information to respond. If appropriate, enter your response in the designated text field and select Submit.

Disable human supervision

To disable human supervision, go to the computer-use tool's Human supervision section and remove all reviewers from the Assign requests to list.

FAQ for the computer use tool