Workload Identity Federation — Azure to GCP Vertex AI

Complete Setup Guide: Keyless Authentication for AI Models on Vertex AI

This guide covers every step required to configure an Azure App Service to call Google Vertex AI (your AI models) using Workload Identity Federation — no stored secrets, no service account keys, keyless machine-to-machine trust.

Overview

Workload Identity Federation (WIF) allows a workload running on Azure (App Service, VM, Container App) to prove its identity to Google Cloud Platform without a long-lived service account key. The flow is:

Azure App Service Managed Identity issues a short-lived Azure AD JWT.

That JWT is exchanged at the GCP Security Token Service (STS) for a federated token.

The federated token is used to impersonate a GCP Service Account.

The service account token is used to call Vertex AI (your model endpoints).

Note: You will need: GCP project owner / editor access, and Azure subscription access to the App Service(s). Steps are split into GCP Admin and Azure Client.

Part 1 — GCP Setup (Admin)

1.1 Enable Required GCP APIs

In the GCP Console for your project, enable the following APIs. All are required — missing any one will cause authentication or model-call failures.

Required: Agent Platform API

Required: Security Token Service (STS) API — enables the WIF token exchange

Required: IAM Credentials API (Service Account impersonation)

Navigation path:
GCP Console → APIs & Services → Library → search each name → Enable

1.2 Create the Workload Identity Pool

A Workload Identity Pool is the container that holds trust relationships with external identity providers (in this case Azure AD).

Navigation:
GCP Console → IAM & Admin → Workload Identity Federation → Create Pool

Fill in the pool form:

Field	Value
Name	e.g. `wif-azure-prod`
Pool ID	Automatically derived from name (lowercase, hyphens). Note this down — used later.
Description	Optional. e.g. "Azure App Service WIF for [your project]"
Enabled	Checked (ON)

Click Continue — do NOT click Done yet; you will add the provider next.

1.3 Add the Azure OIDC Provider

The provider tells GCP to trust JWTs issued by Azure Active Directory. After clicking Continue from the pool form, the Add a Provider screen appears.

Provider settings:

Field	Value
Select a provider	OpenID Connect (OIDC)
Provider Name	e.g. `Azure Provider` (display name only)
Provider ID	e.g. `azure-provider` (URL-safe, lowercase — note this down)
Issuer (URL)	`https://login.microsoftonline.com/{TENANT_ID}/v2.0`
Allowed Audiences	fb60f99c-7a34-4190-8149-302f77469936

1.3a Find Your Tenant ID

In any Azure Portal session:

From the Search page, find Entra ID.

Copy the Tenant ID from the Overview page.

Paste it into the Issuer URL replacing {TENANT_ID}.

Example Issuer URL:

https://login.microsoftonline.com/<tenant_id>/v2.0

Warning: The issuer URL must end in /v2.0. Azure App Service Managed Identity issues v2.0 tokens. Without /v2.0 the token exchange will fail with invalid_grant: issuer mismatch.

1.3b Find the Allowed Audience GUID

You must use the GUID form in the GCP provider Allowed Audiences field.

Enter this Microsoft-provided GUID:

fb60f99c-7a34-4190-8149-302f77469936

1.3c Configure Attribute Mapping

On the same provider form, scroll to Attribute Mapping. These mappings copy claims from the Azure JWT into Google's assertion attributes.

google.subject = assertion.sub

The google.subject mapping is mandatory — it becomes the value used in principal-set bindings (the Object ID / sub claim from Azure).

Click Save to finish creating the pool and provider.

1.4 Create or Identify the GCP Service Account

Vertex AI calls are made by impersonating a GCP Service Account. You can create a dedicated one or reuse an existing account.

Create a new Service Account

Navigation:
GCP Console → IAM & Admin → Service Accounts → + Create Service Account

Field	Value
Service Account Name	e.g. `ejento-service-account`
Service Account ID	Auto-filled. Note the full email — format: `{id}@{project}.iam.gserviceaccount.com`
Description	Optional. e.g. "Used by ejento for Vertex AI calls via WIF"

1.5 Grant Vertex AI Role to the Service Account

The service account must have permission to call Vertex AI model endpoints.

Navigation:
GCP Console → IAM & Admin → IAM → + Grant Access

Fill in:

New principals: paste the service account email

Role: Vertex AI User (roles/aiplatform.user)

Click Save.

Note: Vertex AI User is the minimum required role. Do not grant broader roles (e.g. Editor) unless your use case requires it.

1.6 Grant Workload Identity User to the Service Account

This binding allows the Azure managed identity to impersonate the GCP Service Account. There are two approaches:

Option A — Per App Service (Recommended, Most Secure)

After the customer shares their App Service's Object (Principal) ID, create a specific binding:

Navigation:
IAM & Admin → Service Accounts → select your Service Account → Permissions → Grant Access

New principal (copy exactly — replace placeholders):

principalSet://iam.googleapis.com/projects/{PROJECT_NUMBER}/locations/global/workloadIdentityPools/{POOL_ID}/*

Role: Workload Identity User (roles/iam.workloadIdentityUser)

Option B — Wildcard Binding (broader access)

Alternatively, a wildcard binding can be used to allow any identity within the pool to impersonate the service account, rather than scoping to a single Object ID:

principalSet://iam.googleapis.com/projects/{PROJECT_NUMBER}/locations/global/workloadIdentityPools/{POOL_ID}/*

Role: Workload Identity User (roles/iam.workloadIdentityUser)

Note: This option is broader than Option A since it trusts every identity in the pool, not just one specific App Service. Use Option A when possible for tighter scoping.

3.2 Placeholder Reference Table

Value / Placeholder	Where to Find / Example	Notes
`{PROJECT_NUMBER}`	GCP Console → Home → Project info → Project number	Numeric, not the Project ID string
`{POOL_ID}`	IAM & Admin → Workload Identity Federation → select pool → copy Pool ID e.g. `azure-pool-dev`	Exact ID, not display name
`{PROVIDER_ID}`	Same pool page → Providers tab → Provider ID column e.g. `ejento-azure`	Exact ID, not display name
`{SERVICE_ACCOUNT_EMAIL}`	IAM & Admin → Service Accounts → email column	Full email including `@project.iam.gserviceaccount.com`

Workload Identity Federation — Azure to GCP Vertex AI

Overview#

Part 1 — GCP Setup (Admin)#

1.1 Enable Required GCP APIs#

1.2 Create the Workload Identity Pool#

1.3 Add the Azure OIDC Provider#

1.3a Find Your Tenant ID#

1.3b Find the Allowed Audience GUID#

1.3c Configure Attribute Mapping#

1.4 Create or Identify the GCP Service Account#

1.5 Grant Vertex AI Role to the Service Account#

1.6 Grant Workload Identity User to the Service Account#

Option A — Per App Service (Recommended, Most Secure)#

Option B — Wildcard Binding (broader access)#

3.2 Placeholder Reference Table#