Authorization in Microsoft Entra Agent ID
In brief
The article now distinguishes downstream API permissions from permissions to invoke an agent, clarifies Azure Key Vault role requirements, updates consent guidance, and explains permission handling for multi-instance and single-instance blueprints.
What Entra admins need to know
Administrators can use the clarified guidance to assign narrower resource permissions and avoid confusing API access with agent invocation access.
This editorial summary was generated by AI from the documentation changes. Verify important details in the full Microsoft Learn article.
Documentation change
The comparison below shows only the changed extract. Use the full-page view for complete context.
Authorization in Microsoft Entra Agent ID
The introduction ofAuthorize an agent identities is driven byaccording to the rise of AI-powered agents in organizations. Traditional identity types (like standard app registrations or user accounts) aren't ideal for autonomous agents. AI agents have unique security concerns because their autonomous decision-making, dynamic learning capabilities,resource it needs to access: Azure roles govern Azure resources, Microsoft Entra roles govern directory administration, and potentially accessAPI permissions govern calls to sensitive data can introduce unpredictable behaviors.
Microsoft Graph and other APIs. Microsoft Entra Agent ID was createdrestricts high-privilege assignments to fill this gap. Built on the Microsoft Entra ID platform, it provides a dedicated authentication and authorization framework for AI agents that enables them to safely access services and APIs while giving administrators a central way to monitor and control their actions. In short,limit what an autonomous agent identities let organizations discover, manage, and secure AI agents operating in a tenant, with proper policy enforcement, instead of treating agents as either full users or generic apps.can do.
This article explains how authorization in Microsoft Entra Agent ID worksthose downstream permission controls. Permissions to call a downstream API don't determine who can invoke the agent itself; for AI agents by providing information about roles, permission controls, and best practices for managing agent access.that separate boundary, see Control access to agents with app roles.
Why agent identity authorization is important
Azure roles
If an agent needs to access Azure resources: Assign Azure roles for those specific resources. For example, to let an agent read an Azure Key Vault, give its identity a Key Vault Reader allows reading vault metadata, not secret values. Reading secret values requires a data-plane role such as Key Vault Secrets Useron that vault. This keeps scope narrow (just that resource or resource group) and uses least privilege.a vault configured for Azure RBAC. Assign the role at the narrowest required scope. For more information, see Azure built-in roles for Key Vault and Assign Azure roles using the Azure portal.
Microsoft Entra roles
Delegated Microsoft Graph permissions
If an agent acts on behalf of a user (user-centric scenarios): Use delegated Microsoft Graph permissions. This option requires interactiveThese require user consent but ensuresor administrator consent, and the agent can't exceed thatthe user's access. For example, an agent scheduling meetings for Alice would useuses delegated calendar API permissions; Alice consents, andpermissions. A hosted agent's OBO flow can't display an interactive consent prompt; configure consent before the agent can only manage Alice's calendar (just as Alice could by themselves).downstream call. For more information,supported consent paths, see Overview of permissions and consent in the Microsoft identity platformGrant delegated permissions to an agent identity.
Microsoft Graph application permissions
Inheritable permissions for agent identities
AgentMulti-instance agent identity blueprints support inheritable permissions, which let administrators grant permissions once at theconfigure eligible blueprint level and have those grants automatically apply to allfor use by child agent identities created from the blueprint.identities. This capability reduces repeated consent promptssetup across multiple deployments and environments.a fleet. A single-instance blueprint has no blueprint principal; grant permissions directly to its agent identity instead.
For more information about how inheritable permissions, required resource access, and direct grants work together, see Inheritable permissions. \ No newline at end of file
@@ -1,18 +1,17 @@ --- title: Authorization in Microsoft Entra Agent ID description: Learn about how authorization in Microsoft Entra Agent ID works for AI agents.-ms.date: 06/12/2026+ms.date: 11/05/2026+ai-usage: ai-assisted ms.topic: concept-article #customer intent: As a developer or IT administrator, I want to understand authorization controls for agent IDs so that I can properly configure roles and permissions while maintaining security boundaries for AI agents in my organization. --- # Authorization in Microsoft Entra Agent ID -The introduction of agent identities is driven by the rise of AI-powered agents in organizations. Traditional identity types (like standard app registrations or user accounts) aren't ideal for autonomous agents. AI agents have unique security concerns because their autonomous decision-making, dynamic learning capabilities, and potentially access to sensitive data can introduce unpredictable behaviors.+Authorize an agent according to the resource it needs to access: Azure roles govern Azure resources, Microsoft Entra roles govern directory administration, and API permissions govern calls to Microsoft Graph and other APIs. Microsoft Entra Agent ID restricts high-privilege assignments to limit what an autonomous agent can do. -Microsoft Entra Agent ID was created to fill this gap. Built on the Microsoft Entra ID platform, it provides a dedicated authentication and authorization framework for AI agents that enables them to safely access services and APIs while giving administrators a central way to monitor and control their actions. In short, agent identities let organizations discover, manage, and secure AI agents operating in a tenant, with proper policy enforcement, instead of treating agents as either full users or generic apps.--This article explains how authorization in Microsoft Entra Agent ID works for AI agents by providing information about roles, permission controls, and best practices for managing agent access.+This article explains those downstream permission controls. Permissions to call a downstream API don't determine who can invoke the agent itself; for that separate boundary, see [Control access to agents with app roles](control-user-access-agents.md). ## Why agent identity authorization is important @@ -142,7 +141,7 @@ Depending on what an agent needs to do, administrators can grant access in diffe ### Azure roles -**If an agent needs to access Azure resources**: Assign Azure roles for those specific resources. For example, to let an agent read an Azure Key Vault, give its identity a Key Vault Reader role on that vault. This keeps scope narrow (just that resource or resource group) and uses least privilege. For more information, see [Assign Azure roles using the Azure portal](/azure/role-based-access-control/role-assignments-portal).+**If an agent needs to access Azure resources**: Assign Azure roles for those specific resources. For example, **Key Vault Reader** allows reading vault metadata, not secret values. Reading secret values requires a data-plane role such as **Key Vault Secrets User** on a vault configured for Azure RBAC. Assign the role at the narrowest required scope. For more information, see [Azure built-in roles for Key Vault](/azure/role-based-access-control/built-in-roles/security#key-vault-secrets-user) and [Assign Azure roles using the Azure portal](/azure/role-based-access-control/role-assignments-portal). ### Microsoft Entra roles @@ -150,7 +149,7 @@ Depending on what an agent needs to do, administrators can grant access in diffe ### Delegated Microsoft Graph permissions -**If an agent acts on behalf of a user (user-centric scenarios)**: Use delegated Microsoft Graph permissions. This option requires interactive user consent but ensures the agent can't exceed that user's access. For example, an agent scheduling meetings for Alice would use delegated calendar API permissions; Alice consents, and the agent can only manage Alice's calendar (just as Alice could by themselves). For more information, see [Overview of permissions and consent in the Microsoft identity platform](../identity-platform/permissions-consent-overview.md).+**If an agent acts on behalf of a user (user-centric scenarios)**: Use delegated Microsoft Graph permissions. These require user or administrator consent, and the agent can't exceed the user's access. For example, an agent scheduling meetings for Alice uses delegated calendar API permissions. A hosted agent's OBO flow can't display an interactive consent prompt; configure consent before the downstream call. For supported consent paths, see [Grant delegated permissions to an agent identity](grant-agent-access-microsoft-365.md). ### Microsoft Graph application permissions @@ -160,6 +159,6 @@ Always review the permission's scope: tenant-wide read access might be fine for ## Inheritable permissions for agent identities -Agent identity blueprints support inheritable permissions, which let administrators grant permissions once at the blueprint level and have those grants automatically apply to all agent identities created from the blueprint. This capability reduces repeated consent prompts across multiple deployments and environments.+Multi-instance agent identity blueprints support inheritable permissions, which let administrators configure eligible blueprint grants for use by child agent identities. This capability reduces repeated consent setup across a fleet. A single-instance blueprint has no blueprint principal; grant permissions directly to its agent identity instead. For more information about how inheritable permissions, required resource access, and direct grants work together, see [Inheritable permissions](concept-inheritable-permissions.md).\ No newline at end of file 