Agent identities, service principals, and applications
In brief
The article now distinguishes multi-instance and single-instance Agent IDs, including blueprint principals, client IDs, token flows, credentials, and Microsoft Graph creation guidance.
What Entra admins need to know
Administrators can use the correct setup and authentication guidance for each Agent ID instance mode.
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.
Agent identities, service principals, and applications
Agent identities are built on the same service principal infrastructure that applications use throughout Microsoft Entra, but they're a distinct object type. Standard service principals were designed for static, deterministic workloads and operate using their own credentials directly. Agent identities support one-to-one and one-to-many blueprint relationships, require an assigned sponsor, and generate agent-specific audit entries. Multi-instance blueprints also add a delegation model where the blueprint acquires tokens on behalf of each agent identity, support a one-to-many blueprint relationship, require an assigned sponsor, and generate agent-specific audit entries.Agent ID.
This article explains the following concepts:
- The service principal types in the agent app model
Agent identity blueprint principal
Agent identityA multi-instance blueprint principals are created automatically whenneeds an agent identity blueprint is instantiatedprincipal in a tenant. Theseeach tenant where it operates. Managed creation experiences can create it automatically. When you create the blueprint through Microsoft Graph, create an agent blueprint principal. This service principals provideprincipal provides the runtime representation of the agent identity blueprint within the tenant's directory and enable the agent identityenables it to create Agent IDs and manage their lifecycle. A single-instance blueprint to perform operations, such as creating instances and managing lifecycle operations.doesn't have this principal.
The creation process involves consent operations that require permissions like AgentIdentity.Create and ServicePrincipal.Manage.OwnedBy. The agent identity blueprint principal enables the agent identity blueprint to obtain app-only tokens for Microsoft Graph calls necessary to create and manage agent identities. This process is essential because agent blueprints themselves can't directly obtain tokens for Microsoft Graph operations.
Agent identities are modeled as single-tenant service principals with a new "agent" subtype classification. This design uses existing Microsoft Entra ID service principal infrastructure while adding agent-specific behaviors and constraints.
Agent identities inherit their protocol properties from their parent agent identity blueprint through the ParentID relationship. Unlike standard service principals that operate independently, agent identitiesAgent IDs created from multi-instance blueprints require their parent agent identitythe blueprint for impersonation and token exchange operations. A single-instance Agent ID instead shares the blueprint's OAuth client ID and requests tokens directly. It can operate as a confidential client with a credential on the blueprint or as a public client without a client credential.
Agent identities can be granted permissions directly and appear in sign-in logs when tokens are issued for agent operations. They serve as the primary identity that customers reason about when managing agent permissions and access.
TheFor multi-instance blueprints, the agent identity blueprint principal creates agent identityAgent ID service principals by using Microsoft Graph calls with app-only tokens and appropriate roles. The creation process establishes the parent-child relationship and configures the necessary Federated Identity Credential (FIC)federated identity credential relationships for impersonation.
Agents built using application service principals
Impersonation model
Standard service principals operate using their own credentials and identity. Multi-instance Agent service principalsIDs use an impersonation model: the agent identity blueprint acquires tokens on behalf of each agentchild identity, so the agent identity appears as the client in the resulting token and in audit logs, even though the blueprint performed the actualinitial token exchange.
This separation means the blueprint holds the authentication credentials while the agent identity holds the permissions and the audit identity. A compromise of the blueprint's credentials affects all agent identities under it, which is why blueprint count is a security boundary decision.
A single-instance Agent ID doesn't use this impersonation step. It requests a token with the shared client ID, using a blueprint credential when the flow requires client authentication. For request examples, see Authenticate a single-instance agent identity.
Multi-instance relationship
Standard applications have a one-to-one relationship between application and service principal. TheIn multi-instance mode, the agent app model is one-to-many: a single agent identity blueprint can produce multiple agent identityAgent ID service principals within a tenant or across tenants. Each agent identityAgent ID sources its protocol properties from the blueprint but can hold its own permissions on downstream resources.
In single-instance mode, the relationship is one-to-one in each tenant. The blueprint and its Agent ID service principals share one OAuth client ID, and no blueprint principal is created.
Credential management
Standard service principals manage their own credentials (certificates, secrets, or managed identities) and use them to obtain tokens directly. For confidential Agent identitiesIDs, the blueprint holds the credentials. Multi-instance blueprints use them for impersonation; single-instance Agent IDs use them directly. Public-client Agent IDs don't manage credentials. The blueprint holds all credentials and uses them to impersonate each agent identity.send a client credential.
Managed identities are a supported credential type on blueprints (the most secure option for agents running on Azure), but they're a credential on the blueprint, not a replacement for the agent identity itself.
Permissions and role assignment
Agent service principals support both application permissions (for autonomous operations) and delegated permissions (for interactive, on-behalf-of operations). Permissions can be assigned directly to an agent identity or inheritedAgent ID. Multi-instance Agent IDs can also inherit permissions from the blueprint.blueprint principal.
When InheritDelegatedPermissions is enabled, agent identities inherit delegated permissions from the blueprint, which simplifies consent management across many agent identities. Use inherited permissions for the shared baseline and direct assignment for role-specific access.
@@ -3,15 +3,16 @@ title: Agent identities, service principals, and applications description: Learn about agent service principals in Microsoft Entra Agent ID and how they differ from traditional service principals in authentication, permissions, and lifecycle management. titleSuffix: Microsoft Entra Agent ID ms.topic: concept-article-ms.date: 04/30/2026+ms.date: 11/03/2026 ms.reviewer: dastrock+ai-usage: ai-assisted #customer-intent: As a developer or IT administrator, I want to understand how agent service principals work and how they differ from traditional service principals, so that I can effectively implement authentication, and permissions for agent applications in my organization. --- # Agent identities, service principals, and applications -Agent identities are built on the same service principal infrastructure that applications use throughout Microsoft Entra, but they're a distinct object type. Standard service principals were designed for static, deterministic workloads and operate using their own credentials directly. Agent identities add a delegation model where the blueprint acquires tokens on behalf of each agent identity, support a one-to-many blueprint relationship, require an assigned sponsor, and generate agent-specific audit entries.+Agent identities are built on the same service principal infrastructure that applications use throughout Microsoft Entra, but they're a distinct object type. Standard service principals were designed for static, deterministic workloads and operate using their own credentials directly. Agent identities support one-to-one and one-to-many blueprint relationships, require an assigned sponsor, and generate agent-specific audit entries. Multi-instance blueprints also add a delegation model where the blueprint acquires tokens on behalf of an Agent ID. This article explains the following concepts: - The service principal types in the agent app model@@ -20,7 +21,7 @@ This article explains the following concepts: ## Agent identity blueprint principal -Agent identity blueprint principals are created automatically when an agent identity blueprint is instantiated in a tenant. These service principals provide the runtime *representation* of the agent identity blueprint within the tenant's directory and enable the agent identity blueprint to perform operations, such as creating instances and managing lifecycle operations.+A multi-instance blueprint needs an agent identity blueprint principal in each tenant where it operates. Managed creation experiences can create it automatically. When you create the blueprint through Microsoft Graph, [create an agent blueprint principal](create-blueprint.md#create-an-agent-blueprint-principal). This service principal provides the runtime *representation* of the blueprint within the tenant's directory and enables it to create Agent IDs and manage their lifecycle. A single-instance blueprint doesn't have this principal. The creation process involves consent operations that require permissions like `AgentIdentity.Create` and `ServicePrincipal.Manage.OwnedBy`. The agent identity blueprint principal enables the agent identity blueprint to obtain app-only tokens for Microsoft Graph calls necessary to create and manage agent identities. This process is essential because agent blueprints themselves can't directly obtain tokens for Microsoft Graph operations. @@ -28,11 +29,11 @@ The creation process involves consent operations that require permissions like ` Agent identities are modeled as single-tenant service principals with a new "agent" subtype classification. This design uses existing Microsoft Entra ID service principal infrastructure while adding agent-specific behaviors and constraints. -Agent identities inherit their protocol properties from their parent agent identity blueprint through the ParentID relationship. Unlike standard service principals that operate independently, agent identities require their parent agent identity blueprint for impersonation and token exchange operations.+Agent identities inherit their protocol properties from their parent agent identity blueprint through the ParentID relationship. Agent IDs created from multi-instance blueprints require the blueprint for impersonation and token exchange operations. A single-instance Agent ID instead shares the blueprint's OAuth client ID and requests tokens directly. It can operate as a confidential client with a credential on the blueprint or as a public client without a client credential. Agent identities can be granted permissions directly and appear in sign-in logs when tokens are issued for agent operations. They serve as the primary identity that customers reason about when managing agent permissions and access. -The agent identity blueprint principal creates agent identity service principals using Microsoft Graph calls with app-only tokens and appropriate roles. The creation process establishes the parent-child relationship and configures the necessary Federated Identity Credential (FIC) relationships for impersonation.+For multi-instance blueprints, the agent identity blueprint principal creates Agent ID service principals by using Microsoft Graph calls with app-only tokens and appropriate roles. The creation process establishes the parent-child relationship and configures the necessary federated identity credential relationships for impersonation. ## Agents built using application service principals @@ -46,23 +47,27 @@ The following sections describe how the agent app model differs from standard ap ### Impersonation model -Standard service principals operate using their own credentials and identity. Agent service principals use an impersonation model: the agent identity blueprint acquires tokens on behalf of each agent identity, so the agent identity appears as the client in the resulting token and in audit logs, even though the blueprint performed the actual token exchange.+Standard service principals operate using their own credentials and identity. Multi-instance Agent IDs use an impersonation model: the agent identity blueprint acquires tokens on behalf of each child identity, so the agent identity appears as the client in the resulting token and in audit logs, even though the blueprint performed the initial token exchange. This separation means the blueprint holds the authentication credentials while the agent identity holds the permissions and the audit identity. A compromise of the blueprint's credentials affects all agent identities under it, which is why blueprint count is a security boundary decision. +A single-instance Agent ID doesn't use this impersonation step. It requests a token with the shared client ID, using a blueprint credential when the flow requires client authentication. For request examples, see [Authenticate a single-instance agent identity](authenticate-single-instance-agent.md).+ ### Multi-instance relationship -Standard applications have a one-to-one relationship between application and service principal. The agent app model is one-to-many: a single agent identity blueprint can produce multiple agent identity service principals within a tenant or across tenants. Each agent identity sources its protocol properties from the blueprint but can hold its own permissions on downstream resources.+In multi-instance mode, the agent app model is one-to-many: a single agent identity blueprint can produce multiple Agent ID service principals within a tenant or across tenants. Each Agent ID sources its protocol properties from the blueprint but can hold its own permissions on downstream resources.++In [single-instance mode](agent-blueprint.md#instance-modes-preview), the relationship is one-to-one in each tenant. The blueprint and its Agent ID service principals share one OAuth client ID, and no blueprint principal is created. ### Credential management -Standard service principals manage their own credentials (certificates, secrets, or managed identities) and use them to obtain tokens directly. Agent identities don't manage credentials. The blueprint holds all credentials and uses them to impersonate each agent identity.+Standard service principals manage their own credentials (certificates, secrets, or managed identities) and use them to obtain tokens directly. For confidential Agent IDs, the blueprint holds the credentials. Multi-instance blueprints use them for impersonation; single-instance Agent IDs use them directly. Public-client Agent IDs don't send a client credential. Managed identities are a supported credential type on blueprints (the most secure option for agents running on Azure), but they're a credential *on the blueprint*, not a replacement for the agent identity itself. ### Permissions and role assignment -Agent service principals support both application permissions (for autonomous operations) and delegated permissions (for interactive, on-behalf-of operations). Permissions can be assigned directly to an agent identity or inherited from the blueprint.+Agent service principals support both application permissions (for autonomous operations) and delegated permissions (for interactive, on-behalf-of operations). Permissions can be assigned directly to an Agent ID. Multi-instance Agent IDs can also inherit permissions from the blueprint principal. When `InheritDelegatedPermissions` is enabled, agent identities inherit delegated permissions from the blueprint, which simplifies consent management across many agent identities. Use inherited permissions for the shared baseline and direct assignment for role-specific access. 