Copilot Studio Agent Accounts vs Service Accounts: What IT Leaders Need to Know

TL;DR: A Copilot Studio agent can now get its own Microsoft 365 account, with its own Microsoft Entra identity, mailbox, Teams presence and directory entry. It is a Frontier preview feature, and it trades one set of problems for another. Agent accounts remove passwords and MFA rotation, but today they cannot read documents labelled Confidential or Internal Only, they do not respond to anyone outside your tenant, they keep one shared memory across everyone they talk to, and they need E5 plus Agent 365 (or E7) for the maker and a Microsoft 365 Frontier for AI Teammates license for each agent. If your agent talks to customers on Gmail or needs Confidential SharePoint content, keep it on a service account for now and lock that account down.

Microsoft has published new guidance, Give your agent an account (Frontier), that lets makers give a Copilot Studio agent its own identity in Microsoft 365. It is the clearest sign yet that agents are becoming members of the organization rather than features inside an app.

There is no retirement date forcing your hand here, and that is exactly why you need to set your own. When a capability like this lands, someone on the team will propose deleting every agent service account to clean things up. Before that conversation happens, decide which agents move, which stay and how you will govern both. This article gives you the comparison, the limits Microsoft documents today and a plan you can run this quarter.

What is changing with Copilot Studio agent accounts?

Until now, most organizations that wanted an agent to send email, sit in Teams or work on files did it with a service account: a regular user account, a password, MFA and a license, shared by whoever built the agent. An agent account replaces that pattern with an identity that Microsoft issues and governs. When you publish the agent, Copilot Studio provisions a mailbox, Teams presence, directory entry and Microsoft Entra Agent IDs, and the account is marked in the directory as an AI agent that reports to its maker.

Here is how the two options compare today.

What is not changing: agents that only call tools, run in the background or act on behalf of a signed-in user do not need a user account at all. Microsoft's own guidance is to add one only when the agent genuinely needs a mailbox, Teams presence or other user-based Microsoft 365 experiences. For a deeper look at who needs which license to govern agents, read Agent Identity Governance Licensing: Why Your E3 and E5 Tenants Cannot Govern Agents.

What else changes beyond passwords and labels?

The headline comparison is only part of the story. These details from Microsoft's documentation will shape your design more than the headline does.

• Shared memory across every user. An agent with an account keeps one memory for everyone it works with, even in one-to-one chats. Microsoft states that files and chats people share can be summarized and used in responses to any user, even users without permission to the original content, regardless of sensitivity label. End users cannot opt out in this preview, and admins cannot turn shared memory off without turning off agent accounts altogether. Memories are deleted automatically after 28 days with no interaction.
• A 250-person email ceiling. The agent responds only in email conversations with up to 250 people. Large distribution lists are out.
• New documents, not edits. The agent can generate new documents but cannot co-author or make targeted edits to existing ones yet.
• Partial Microsoft Graph coverage. Not every Graph operation or permission scope is supported, so test each tool the agent depends on.
• Identity decisions are permanent. The email alias uses the maker's alias as a fixed prefix, the agent reports to the maker, and neither the prefix, the reporting line nor the email address can change after you publish.
• Specific setup requirements. Your environment must use the early release cycle, makers work in Copilot Studio Preview, an admin must activate the Copilot Studio (Frontier) agent template for selected users and turn on Agent accounts in the Power Platform admin center, and the account must be added before the agent is published for the first time.
• The GitHub Copilot harness. Agent accounts are used by agents powered by the GitHub Copilot harness, with usage-based billing in Copilot Credits for building, testing and running. If you are still deciding on a runtime, see Harness Choice Is a Design Decision, Not a Toggle.
• Preview terms apply. Microsoft notes there is no specific service level agreement for Agent 365 Frontier, and capabilities can change or be removed.

Why do agent accounts matter for CIOs, CHROs and IT leaders?

Service accounts were always a workaround. They look like people in your audit logs, their passwords end up in a password manager that three makers share, and when the builder leaves, nobody is sure who owns the account. Agent accounts fix that problem at the root. Every agent becomes a visible, sponsored identity that you can govern with Conditional Access and Agent 365 instead of with password policies that were designed for humans.

The trade-off is that the new model enforces your data protection more strictly than your old workaround did. That is good news for security and a surprise for anyone whose agent quietly depended on reading a Confidential site. Leaders should expect three conversations:

• Security and compliance will welcome the end of shared passwords and will want a clear position on shared memory before any pilot.
• HR should weigh in early. An agent that sits in Teams, has a job title and reports to a person changes how employees perceive the work, and shared memory means anything shared with the agent may resurface for colleagues.
• Finance will ask what each agent costs. Frontier licensing, maker licensing and Copilot Credits are separate lines, and the model may change after preview. Microsoft 365 Copilot Licensing Changes: December 2026 Guide covers the usage-based billing shift that sits alongside this.

Some sectors have specific considerations:

• Nonprofits often run agents that talk to donors, volunteers and community members on personal email. Those conversations are outside the tenant, so they stay on a service account for now. Also check that your licensing path includes the E5 plus Agent 365 or E7 maker license before you plan a pilot.
• Education institutions should keep agents that serve parents, applicants or alumni on service accounts, since those people are usually external. For internal agents, think carefully about shared memory near student records: a summary shared with the agent by one staff member could surface in a reply to another.
• Government organizations should confirm Frontier and Agent 365 availability in their own cloud before planning, since several Agent 365 capabilities are not yet available in GCC. Records retention and access-to-information obligations also need a clear answer for content the agent remembers.

What should IT leaders do before replacing service accounts?

1. Inventory every agent service account. List each account an agent or flow uses, its owner, the sites and mailboxes it can reach, whether it talks to external people and which sensitivity labels sit on its content. If you cannot count your agents yet, start with Your Agent Estate Doubled. Your Controls Didn't.
2. Sort each agent into one of three groups. Keep it on a service account if it needs external contacts, protected content, edits to existing documents or audiences over 250. Mark it as an agent account candidate if it runs an internal team process on unprotected content. Mark it as needing no user account if it only calls tools, runs in the background or acts on behalf of a signed-in user.
3. Harden the service accounts you keep. Apply least access, strong phishing-resistant MFA through Conditional Access, block interactive sign-in where you can, assign a named owner and a backup, and review access every quarter.
4. Decide your shared memory position before any pilot. Agree which teams can use an agent account, which content may be shared with it and how you will tell users that what they share can resurface for others.
5. Set up the Frontier prerequisites deliberately. Have a global admin enroll the tenant and confirm Microsoft 365 Frontier for AI Teammates licenses, activate the agent template for a small maker group only, and turn on Agent accounts in a single early-release environment. Agree the agent's name and alias before publishing, because they cannot change later.
6. Pilot one internal process with approved samples. Microsoft's own example is a team execution-review agent that reads a tracker, writes a new review document and asks reviewers for context. Use non-sensitive test content, and never relabel documents to make the agent work.
7. Bring every agent identity into your operating model. Give each one a sponsor, a lifecycle and a review cadence, and restrict who can share agents widely using New Copilot Studio Sharing Controls Every Organization Should Enable. An Agent PMO makes this routine, as described in The Agent PMO Explained for Copilot Governance.

How do you tie agent account spend to measurable value?

An agent account carries real cost: the maker's E5 plus Agent 365 or E7 license, one Microsoft 365 Frontier for AI Teammates license per agent, Copilot Credits for usage and the admin time to run a preview. A service account has costs too, just less visible ones: a license, credential rotation, help desk time when MFA breaks and the risk carried by a shared password.

Compare both on one simple scorecard:

• Identity overhead: hours a month spent rotating credentials, fixing MFA and chasing ownership
• Risk: accounts with no named owner, accounts with access beyond their task and sign-in alerts per quarter
• Outcome: cycle time for the process the agent runs, items it handles and how often a person has to step in
• Cost: licenses, Copilot Credits consumed and maker hours to maintain the agent

For example, an execution-review agent that saves a program manager three hours a week returns roughly 150 hours a year. If the same agent also retires a shared service account that needed monthly attention, that is time and risk you can put a number on. If an agent account cannot do the job because of a documented limit, record that too. It tells you exactly what to retest when the preview changes.

For the broader picture of what E7 bundles and when it pays off, read Microsoft 365 E7 (The Frontier Suite): What You Get and Why Agents Change Everything. To turn usage signals into an ROI story leadership will fund, see Beyond 'Active Users': The Copilot Adoption Metrics That Actually Prove Value.

Copilot Studio agent accounts FAQ

Should we delete our agent service accounts now?

No. Agent accounts are a Frontier preview with documented limits. Move only the agents that fit those limits, and keep the rest on service accounts that have least access, strong MFA and a named owner.

Can an agent account read Confidential files if we grant it access?

No. Microsoft states that agents with an account cannot read protected documents, such as files labelled Confidential or Internal Only. Do not downgrade labels to make an agent work. Keep agents that need that content on a governed service account for now.

Can customers email an agent account from Gmail or Outlook.com?

No. In this preview the agent does not respond to anyone outside its tenant, and Microsoft says there is no workaround. Customer-facing and donor-facing agents should stay on a service account.

Who can see what an agent account remembers?

Anyone who works with the agent. Shared memory can surface summaries of files and chats to other users, even users without access to the original content, regardless of sensitivity label. Users cannot opt out, and admins can only turn it off by turning off agent accounts.

How can 2toLead help?

2toLead helps small and mid-market organizations across healthcare, nonprofit, education, financial services, professional services, government and manufacturing build, govern and measure Copilot Studio agents. Before you change how your agents sign in, we can help you:

• Inventory agent service accounts, confirm owners and sort each agent into keep, move or no account needed
• Harden the service accounts you keep with least access, Conditional Access and access reviews
• Set up a safe Frontier pilot, including licensing checks, environment setup and a shared memory policy your security and HR leaders agree on
• Put a scorecard in place so every agent identity has a cost, an owner and a measurable outcome

For more on Copilot, agents and governance, browse the 2toLead Insights hub. When you are ready, book an agent identity review with our team before anyone deletes a service account.

No hype. Just practical, governed AI that actually gets adopted.