What Entra ID does well
Microsoft Entra ID, formerly Azure Active Directory, is the identity service behind Microsoft 365. For most organisations on Microsoft 365, it already knows who works there, which groups they belong to and which applications they may use. That makes it the natural foundation for AI access: you do not want a separate list of people for every AI tool.
Four practical uses
1. Single sign-on to AI tools
Configure your AI tools and any AI gateway to sign people in through Entra ID, where the tool supports it. When someone leaves and their account is disabled, their access to those tools goes with it.
2. Control app consent
AI tools and connectors often ask for permission to read mail, files or calendars. Entra ID lets administrators decide whether users can consent to applications themselves, or whether an administrator must approve. For AI connectors, requiring administrator review of broad permissions is a sensible default. Check your tenant’s current consent settings.
3. Groups for access decisions
Use Entra groups to decide who may use which AI capability, for example “Tax team”, “Partners” or “AI pilot”. Groups are easier to review than lists of individual grants, and they follow people as they change roles.
4. Central review and revocation
Review enterprise applications and their permissions regularly. Look for AI tools with application-level access to mail or files across the tenant; these deserve particular scrutiny.
Delegated and application permissions
Entra distinguishes between two kinds of permission:
- Delegated permissions let an application act as the signed-in person, limited to what that person can access.
- Application permissions let an application act as itself, often across the whole tenant, without a signed-in person.
For AI working on someone’s behalf, delegated permissions usually fit better, because the AI cannot see more than the person. Application permissions suit background processes but concentrate a great deal of access in one place.
What Entra does not decide
Entra tells you who someone is and which applications they may use. It does not, by itself, decide what a specific AI agent may do on a specific client’s data this afternoon. Questions like these sit above identity:
- Which agent is this, and has the person authorised it for this task?
- May this agent create a draft but not send it?
- May it see this client but not that one?
- Does this action need a partner’s approval?
Those are delegation and policy questions. They need a layer that understands both the person and the agent.
A starter checklist
- Put AI tools behind Entra single sign-on where supported.
- Restrict user consent for applications requesting broad permissions.
- Create groups for AI access and review them regularly.
- Audit enterprise applications for AI connectors with tenant-wide access.
- Decide where agent-level rules will be enforced.
How Kroy approaches it
Kroy can use Entra ID as the source of identity, so people sign in with the accounts and groups you already manage. On top of that identity, Kroy identifies the agent, the delegation between the person and the agent, and your policies, and applies them to every request. Disable someone in Entra and their AI access through Kroy stops with it.