AI Agents Are Getting Their Own User Accounts – What to Settle in Microsoft 365 Before 4 November

AI agents are becoming digital employees with their own account, their own email address and their own permissions. That raises a question that used to apply only to people: who gives an agent access – and who takes it away again?

Two announcements this month make this concrete. On 8 October 2026, Google unveiled a Gemini agent that gets its own Workspace account as a «coworker». And on 13 October 2026, Microsoft announced two changes to Microsoft Entra that will roll out to every Microsoft 365 tenant from 4 November 2026. One of them affects every company using Microsoft 365 – even one that has not deployed a single agent yet.

From tool to team member

Until now, AI assistants worked on behalf of the person using them. Whoever asked a question lent the AI their own permissions for that moment. For agents that work independently on a task for days, that model no longer fits.

Google has therefore taken a different route. According to the launch coverage, such an agent gets its own Workspace account with an email address, calendar, Drive storage and an entry in the company directory. It acts under its own identity rather than the user's, and sees only what is shared with it. For now it is available in private preview to selected Gemini Enterprise customers.

With Microsoft Entra Agent ID, Microsoft is building the same principle into Microsoft 365's identity management. Agents get their own identities there, created from a so-called «agent identity blueprint». According to Microsoft's documentation, every blueprint and every agent identity needs at least one sponsor: an accountable business owner who decides on the agent's purpose, access reviews and lifetime. Technical owners, who manage settings and credentials, can be added as well.

At heart, this is a good development. An agent with its own account is traceable: you can see what it did, grant and revoke its rights precisely, and switch it off without locking a person out.

What changes in Microsoft 365 on 4 November

On 13 October 2026, Microsoft published two changes in the Message Center. Both roll out worldwide from 4 November 2026 and are expected to be complete by late November.

• Ordinary users may create blueprints (MC1506733): People without an admin role will be able to create new agent blueprints – unless an administrator switches the setting off. Existing blueprints are not affected. The setting can be configured ahead of the date.
• Agents may no longer multiply on their own (MC1506728): Until now, blueprints were automatically allowed to create their own agent identities. From 4 November, that requires explicit administrator consent. Anyone running agents that create their own identities must grant this permission beforehand, or that step will fail.

The second change tightens things and mainly affects companies already building their own agents. The first one loosens things, and it affects everyone. If you do nothing, from November essentially anyone in your company can lay the groundwork for new agents.

Why this is tricky for SMEs

With the relaxation, Microsoft wants to lower the barrier to building agents. That is understandable. In practice, though, a familiar pattern repeats itself: anyone can create something, nobody feels responsible, and two years later no one remembers what an account is for or what it may do.

Most SMEs know this problem from user accounts. When someone leaves, their account is – hopefully – properly removed. For agents, that process usually does not exist yet. Yet an agent can do more damage than a forgotten employee account: it works around the clock, it acts on its own, and it does exactly what it is permitted to do – even when nobody wants that any more.

Then there is liability. If an agent sends an email, places an order or shares customer data on your company's behalf, legally that is your company acting. An agent identity makes this traceable. Without clear accountability, it does not help much.

Our view

We see dedicated identities for agents as the right approach, whether an agent runs on Claude, ChatGPT or another model. Agents that run with a person's permissions are hard to control and barely traceable. The question is not whether your company will have agent accounts, but whether they are planned or simply accumulate.

So treat agents like new hires. They get a job description, an accountable person, only the rights they need for their task, and an end date or a regular review. Microsoft's sponsor requirement points exactly this way. But it only works if the sponsor knows they are one.

For most SMEs, we recommend restricting the creation of agent blueprints to administrators for now. That holds back nobody with a good use case: they register it, and the agent is set up deliberately. We described what guardrails for agents can look like in our post on OpenAI's training pause.

What to do now

• Check the setting before 4 November: Have the Microsoft Entra user settings configured to decide whether ordinary users may create agent blueprints. Without a deliberate decision, the open option applies from November.
• Take stock of existing agents: Ask your IT team or provider whether agents or agent blueprints already exist in your tenant – and whether any of your own agents rely on creating identities themselves.
• Name a sponsor for every agent: Someone from the business, not from IT, who can explain what the agent is for and whether it is still needed.
• Keep permissions small: An agent gets access to the folders and mailboxes it needs for its task – not to everything that is technically possible.
• Find out where you stand: Our free AI Readiness Check shows in a few minutes where your company stands on rules, data and accountability. If you are planning agents, we help you build them with our AI services.