Skip to content

Knowledge base

Copilot Studio agents: who can do what, with whose rights?

By Bob Biesemans · published on

A Copilot Studio agent can read documents and take actions. Whose permissions it uses decides what colleagues can do with it. What an SME should watch for.

Short answer

A Copilot Studio agent can run with the maker's credentials or the user's. With the maker's, a colleague can see or do things only the maker may. Choose end-user permissions where you can, give an agent only the sources and actions it needs, and agree who may build and share agents.

An agent does more than answer

With Copilot Studio, employees build their own agents: an assistant that answers from company documents, or one that takes actions through connectors, such as creating a record or sending an email. That is powerful, which is exactly why it matters which data the agent sees and what it may do.

Copilot in Microsoft 365 works with what the user can see. With an agent it depends on how it was built: which knowledge sources it has, which connectors it uses and which credentials those connectors run with.

The maker's rights or the user's

Whoever builds an agent can choose to use their own credentials for a connector. Microsoft explains that the agent then uses the maker's credentials every time it runs, not those of the person asking. A colleague could retrieve information or perform actions that only the maker's account is allowed to.

With end-user credentials, each user signs in to the connected service and only sees what their own account allows. By default both options are enabled. An admin can set, per environment in the Power Platform admin center, that only end-user credentials are allowed. Note that agents running without an active user, for example on a schedule, then stop working.

What an SME should watch for

Agree who may build and share agents, and with whom. An agent shared with the whole company deserves a second look before it goes live. Give an agent only the knowledge sources and actions its task needs, and choose end-user permissions where you can.

Look at the sources too. If a folder of contracts or salary data is shared too widely, an agent using it as a source can surface that information. And keep a list of agents in use, who owns each one and when it was last reviewed.

The skill behind the settings

Settings help, but the choices are made by the employee who builds the agent. Do they understand the difference between their own permissions and the user's? Do they recognise a source that doesn't belong in an agent? That is a different skill from using Copilot every day, so it is best measured separately, with tasks that resemble real building work. Record what you measure and improve as an AI literacy measure under Article 4 of the AI Act.

Source: Microsoft Learn, "Control maker-provided credentials for authentication" (Copilot Studio)

Frequently asked questions

Does an agent only see what I can see?

Not always. If a connector uses the maker's credentials, the agent runs with the maker's permissions. With end-user credentials, each user only sees what their own account allows.

Can an admin block maker credentials?

Yes. In the Power Platform admin center, an admin can set per environment that only end-user credentials are allowed. Agents that run without an active user then stop working.

What do you measure in employees who build agents?

Whether they understand the difference between their own permissions and the user's, recognise a source that doesn't belong in an agent, and limit an agent to what the task needs. Measure that separately from everyday Copilot use.

Further reading

Read next

Help us improve the website?

With your permission, Google Analytics, Microsoft Clarity and Vercel Web Analytics measure visits to our public pages. Your test answers, results and dashboard are not sent. We do not use advertising cookies. Privacy policy