Azure — Discovery

Connect your Azure tenant so SecureAI can inventory your AI resources in Azure and every non-human identity (NHI) in Microsoft Entra ID, with the roles and permissions each one holds.

This connection only reads and never sends a prompt. To run SecureAI’s LLM traffic on your own Foundry resource instead, see Azure AI Foundry — Inference in your account. They are two independent connections with different service principals and permissions; you can use both on the same subscription.

What SecureAI imports

For each subscription the service principal can read:

AI resources: every Azure AI Services, Azure OpenAI and other Cognitive Services account (Content Safety, Speech, Language, Document Intelligence…), with its region and state.
Model deployments of Azure OpenAI and Foundry, with their capacity and the last 30 days of usage (requests, input and output tokens, from Azure Monitor).
Foundry projects and agents: the Foundry Agent Service agents and the classic assistants of every project, with the model each one uses. Also hubs and hub-based projects (Azure Machine Learning), with their online and serverless endpoints.
Azure AI Search and Azure Bot Service.
Cost for the current month, by service (Cost Management).
Activity: Cognitive Services and Machine Learning management events from the last 30 days (Activity Log).
NHIs, listed in the NHI Inventory:
- Service principals of enterprise applications.
- Microsoft Entra agent identities, the ones Foundry creates for each agent (classified as agent).
- Managed identities, system- and user-assigned.
- App registration client secrets and certificates, with their expiry.
- Federated credentials (workload identity federation). For example, a GitHub Actions OIDC trust lets that repository get tokens as the application, with no secret at all.
- The last sign-in of each application, when your tenant exposes that report.
Permissions of each identity: its Azure RBAC roles at any scope of the subscription (subscription, resource group or resource), its Microsoft Graph application permissions and its Entra directory roles.
RBAC role assignments: each role by name, with the users, groups and applications that hold it.

An identity is flagged as privileged when it holds any of these:

one of the Owner, Contributor, User Access Administrator or Role Based Access Control Administrator roles;
a Graph write permission, such as Application.ReadWrite.All, Directory.ReadWrite.All, Mail.Send or Files.ReadWrite.All;
a directory admin role, such as Global Administrator or Application Administrator.

The secrets, certificates and federated credentials of a privileged application are flagged as privileged too. SecureAI does not import the ~130 service principals Microsoft creates in every tenant (Azure Portal, Microsoft Graph, etc.) nor social sign-in connections: they are not your identities to govern.

Prerequisites

A Microsoft Entra ID app registration (service principal) with a client secret and these permissions:

Where	Permission	Required?	What it enables
Subscription	Reader role	Yes	Resources, deployments, projects, managed identities, role assignments, usage and activity
Subscription	Cost Management Reader role	Recommended	Month-to-date cost by service
Foundry resource	Cognitive Services Data Reader role	Recommended	The agents of each Foundry project
Microsoft Graph	`Application.Read.All` (application permission)	Yes	Service principals, agent identities, applications, secrets, certificates, federated credentials and Graph permissions
Microsoft Graph	`Directory.Read.All` (application permission)	Yes	Directory roles and the names of each role’s members
Microsoft Graph	`AuditLog.Read.All` (application permission)	Recommended	Last sign-in of each application
Microsoft Graph	`Application.ReadWrite.All` (application permission)	Only to revoke	Disable service principals and delete secrets, certificates and federated credentials from SecureAI

The Graph permissions need admin consent. To read all your subscriptions, assign Reader on each one, or on a management group that contains them.

Foundry agents live in the project’s data plane, not in Azure Resource Manager: Reader is not enough to see them. Without Cognitive Services Data Reader (or Foundry User) on the Foundry resource, the sync still works, but the agent list stays empty and SecureAI reports it as a gap. Cognitive Services Data Reader is the narrowest role that allows it: data-plane read only.

Step 1. Create the service principal

With the Azure CLI

From Azure Cloud Shell or the az CLI. Replace SUB_ID with your subscription and, if you use Foundry, MY_RESOURCE and MY_RG with your resource and resource group:

SUB_ID=00000000-0000-0000-0000-000000000000

# 1. App registration + service principal
APP_ID=$(az ad app create --display-name "SecureAI Discovery" --query appId -o tsv)
az ad sp create --id "$APP_ID"

# 2. Client secret (shown only once: copy it now)
az ad app credential reset --id "$APP_ID" --display-name secureai --years 1 \
  --query "{tenantId:tenant, clientId:appId, clientSecret:password}" -o json

# 3. Microsoft Graph application permissions + admin consent
#    Application.Read.All, Directory.Read.All, AuditLog.Read.All
az ad app permission add --id "$APP_ID" --api 00000003-0000-0000-c000-000000000000 \
  --api-permissions 9a5d68dd-52b0-4cc2-bd40-abcf44ac3a30=Role \
                    7ab1d382-f21e-4acd-a863-ba3e13f7da61=Role \
                    b0afded3-3588-46d8-8b3d-9842eff778da=Role
az ad app permission admin-consent --id "$APP_ID"

# 4. Azure RBAC on the subscription
az role assignment create --assignee "$APP_ID" --role "Reader" --scope "/subscriptions/$SUB_ID"
az role assignment create --assignee "$APP_ID" --role "Cost Management Reader" --scope "/subscriptions/$SUB_ID"

# 5. (Recommended) read the agents of your Foundry resource
RID=$(az cognitiveservices account show -n MY_RESOURCE -g MY_RG --query id -o tsv)
az role assignment create --assignee "$APP_ID" --role "Cognitive Services Data Reader" --scope "$RID"

Step 2 prints the three values you will paste into SecureAI. To revoke from SecureAI, also add 1bfefb4e-e0b5-418b-a88f-73c46d2cc8e9=Role (Application.ReadWrite.All) in step 3 and grant consent again.

With the Azure portal

In the Azure portal, go to Microsoft Entra ID → App registrations → New registration, give it a name (for example SecureAI Discovery) and register it.
In Overview, copy the Directory (tenant) ID and the Application (client) ID.
In Certificates & secrets → Client secrets → New client secret, create one and copy its Value (it is shown only once).
In API permissions → Add a permission → Microsoft Graph → Application permissions, add Application.Read.All, Directory.Read.All and AuditLog.Read.All (and Application.ReadWrite.All if you want to revoke from SecureAI). Then click Grant admin consent.
In Subscriptions → your subscription → Access control (IAM) → Add role assignment, assign Reader and Cost Management Reader to the application.
In your Foundry resource → Access control (IAM), assign Cognitive Services Data Reader to the application.

Azure roles can take up to 5 minutes to apply, and Graph consent a few minutes more. If the first sync shows gaps for permissions you just granted, wait and sync again.

Treat the secret like a password. SecureAI stores it encrypted and never shows it again. Secrets expire (after one year with the command above): rotate it before then, or the sync stops working that day.

Step 2. Connect Azure in SecureAI

Go to Admin → Integrations, filter by Cloud and, on the Azure — Discovery card, click Connect Integration.

Fill in the fields:

Field	What to enter
Azure Tenant ID (Directory ID)	`tenantId` from step 1
App Registration Client ID	`clientId` from step 1
App Registration Client Secret	`clientSecret` from step 1. Stored encrypted.
Subscription ID (optional)	Limits the read to that subscription. Empty: every subscription where the principal has Reader.

Click Connect. Before saving, SecureAI tests the credentials with a real read: if the secret is wrong or the principal cannot read the subscription, nothing is stored and the message says what is missing. If it connects but a Graph permission is missing, the notice names the permission.
Sync with the card’s refresh button. An Azure sync can take several minutes (it walks Azure Resource Manager, Microsoft Graph, Azure Monitor and Cost Management). After that, SecureAI syncs on its own every 6 hours.

Step 3. Verify

Cloud Sensors lists the discovered deployments, agents and services. Click the Azure card to open its Insights:
- Usage & Costs: usage and cost;
- Models & Deployments: deployments and agents;
- Identity & Access: identities with their roles and the Privileged flag; secrets, certificates and federated credentials; RBAC roles with their members;
- Audit & Logs: the activity.
The NHI Inventory shows service principals, agent identities, managed identities, secrets, certificates and federated credentials, with their permissions and whether they are privileged.

If a sync fails (for example, because the secret expired), the card shows the error and SecureAI retries with backoff, keeping the last inventory.

Coverage gaps

Nothing SecureAI could not read is skipped silently: when you connect or sync, the notice names the missing permission.

Surface	Permission that fixes it
Entra identities (service principals, app registrations)	`Application.Read.All` with admin consent
Graph permissions of each identity	`Application.Read.All`
Directory roles	`Directory.Read.All`
Role assignments and managed identities	Reader on the subscription
Foundry agents	Cognitive Services Data Reader on the Foundry resource

A partial read (a missing permission, a page Azure did not return, throttling) never retires agents or deletes identities from the inventory: SecureAI only treats something as deleted when the listing that would show it answered in full. Without role assignments, an identity’s permissions stay unknown, never “none”.

Revocation

From the NHI Inventory (padlock), with Application.ReadWrite.All:

Identity	What happens in Azure	Reversible?
Service principal	Disabled (`accountEnabled: false`)	Yes: enable it again in Entra ID
Client secret	Deleted from the application	No: create a new one
Certificate	Removed from the application	No: upload it again
Federated credential	The trust is deleted	No: create it again
User-assigned managed identity	The resource is deleted (also needs Managed Identity Contributor on it)	No

Safeguards:

SecureAI does not offer to revoke the service principal it connects with, nor its secrets, because that would cut discovery. Rotate them in Entra ID.
System-assigned managed identities are not revoked from SecureAI: their lifecycle is the Azure resource’s.
Federated credentials that Foundry keeps on an agent’s identity blueprint are not revoked from SecureAI either: cutting them stops the agent.
Automatic NHI policies (revocation by age or disuse) do not apply to Azure yet: revocation is manual.

Disconnect

In Admin → Integrations → Cloud → Azure — Discovery, deleting the connection removes the stored credentials. To remove SecureAI’s access on the Azure side, delete the app registration or its secret.

Getting Started

Security, SMLTP & DLP

AI Gateway

Identity & Access (SSO)

AI Discovery (Shadow AI)

Endpoint Agent

Integrations: Cloud AI & IDEs

Integrations: Files (Drive & SharePoint)

Integrations: SIEMs, Network & MCP

Azure — Discovery

Azure — Discovery

What SecureAI imports

Prerequisites

Step 1. Create the service principal

With the Azure CLI

With the Azure portal

Step 2. Connect Azure in SecureAI

Step 3. Verify

Coverage gaps

Revocation

Disconnect

​Azure — Discovery

​What SecureAI imports

​Prerequisites

​Step 1. Create the service principal

​With the Azure CLI

​With the Azure portal

​Step 2. Connect Azure in SecureAI

​Step 3. Verify

​Coverage gaps

​Revocation

​Disconnect

​Related

Azure — Discovery

What SecureAI imports

Prerequisites

Step 1. Create the service principal

With the Azure CLI

With the Azure portal

Step 2. Connect Azure in SecureAI

Step 3. Verify

Coverage gaps

Revocation

Disconnect

Related