Third-party Registration and Integration
Register AI Red Teaming as a trusted application in a third-party identity provider
to authenticate and run security scans against protected APIs and AI agents.
| Where Can I Use This? | What Do I Need? |
|---|
- Prisma AIRS (AI Red Teaming)
|
- Prisma AIRS AI Red Teaming License
- Prisma AIRS AI Red Teaming Deployment Profile
- (For Entra ID scenarios) An Azure account with an
active subscription
- (For Entra ID scenarios) Admin access to Microsoft
Entra ID (to grant application permissions)
- (For n8n) An active n8n instance (cloud
or self-hosted) with a published workflow and production webhook
URL
- (For Gemini Agent Studio)
A GCP project with a Reasoning Engine agent deployed, with
permission to create custom roles, service accounts, and manage
IAM
|
Many AI systems and agents restrict access behind authentication layers. To run security
scans against these protected targets, AI Red Teaming must authenticate as a trusted
client before it can send attack payloads. Depending on the target, this means
registering AI Red Teaming as an application in a third-party identity provider, or
configuring webhook-level credentials that the target's endpoint validates on each
request.
Supported Authentication Scenarios
AI Red Teaming supports authentication for the following third-party integration
scenarios:
- OAuth 2.0 client credentials flow: Use this when your target is a REST
API or streaming endpoint protected by Azure Entra ID. This flow is designed for
server-to-server, fully automated authentication. AI Red Teaming uses the
client_credentials grant type to fetch and cache access
tokens throughout the scan without requiring a user to be present. Token refresh
is handled automatically.
- Microsoft Copilot Studio: Use this when your target is a Microsoft
Copilot Studio agent. This scenario uses a one-time delegated authentication
flow where you log in through Microsoft once to authorize AI Red Teaming. This
generates a refresh token valid for 90 days that AI Red Teaming uses to
authenticate for subsequent scans.
- n8n: Use this when your target is an AI agent built in
n8n. n8n agents expose an authenticated webhook URL that AI Red Teaming calls
directly. No identity provider registration is required. Authentication uses
either a custom HTTP header (Header Auth) or standard
credentials (Basic Auth) configured on the n8n Webhook
node.
- Gemini Agent Studio: Use this when your
target is an AI agent built in Gemini Agent Studio (Agent Platform). This
scenario requires no OAuth 2.0 registration or identity provider. You create a
custom IAM role and service account in your GCP project, then grant AI Red
Teaming's connector service account the right to impersonate it. No credentials
are downloaded or shared. Authentication uses GCP IAM role grants
only.
| Scenario | Target Type | Auth Mechanism | User Interaction Required |
|---|
| OAuth 2.0 client credentials | REST APIs or streaming endpoints protected by Entra ID | client_credentials grant | None (fully automated) |
| Microsoft Copilot Studio | Copilot Studio agents | authorization_code grant (one-time) | Required once for initial authorization |
| n8n | n8n workflow agents (Agent target type) | Header Auth or Basic Auth on the n8n Webhook node | None (fully automated) |
| Gemini Agent Studio | Gemini Agent Studio agents (Reasoning Engine deployments) | GCP IAM impersonation (Service Account Token Creator) | None (fully automated) |
What Registration Enables
For both Microsoft Entra ID scenarios, you register AI Red Teaming as an application
in Microsoft Entra ID. The registration process differs depending on your
target:
- (For OAuth 2.0) Configure Application permissions (not Delegated) and
skip the redirect URI. AI Red Teaming authenticates using its own identity and
does not require a user sign-in during scanning. Admin consent is required to
activate the Application permissions.
- (For Microsoft Copilot Studio) Configure Delegated permissions with
the CopilotStudio.Copilots.Invoke scope and add a redirect
URI. The redirect URI receives the authorization code after your one-time login,
which AI Red Teaming exchanges for a refresh token.
n8n Integration
n8n is a workflow automation platform for building AI agents. An n8n agent typically
chains a Webhook trigger node, an AI model node, and a Respond to Webhook node. AI
Red Teaming connects to the agent through its published webhook URL, sending attack
prompts as HTTP POST requests and reading the AI model's responses.
n8n integration does not require application registration with an identity provider.
You configure authentication directly on the n8n Webhook node and provide the same
credentials when creating the target in AI Red Teaming. The n8n connection method
supports REST and streaming response modes, single-turn and multi-turn
conversations, and multimodal file attacks against n8n agents. To configure an n8n
agent as a target, follow the steps in
n8n Connection Method. For
authentication options, response modes, multi-turn behavior, and file attack
details, see
n8n and AI Red Teaming Integration.
Gemini Agent Studio Integration
Gemini Agent Studio (previously Vertex AI Agent Studio, now also called Agent
Platform) lets you build no-code and low-code AI agents that run as Reasoning
Engine deployments in your Google Cloud project. AI Red Teaming connects to these
agents through the Vertex AI API using GCP service account impersonation — you
create a dedicated service account and custom IAM role in your project, then grant
AI Red Teaming's connector service account the right to impersonate it. No OAuth
2.0 registration, redirect URIs, or credentials are downloaded or shared.
