Azure Guardrails

Integrate Microsoft Azure’s powerful content moderation services & PII guardrails with Portkey

Microsoft Azure offers robust content moderation and PII redaction services that can now be seamlessly integrated with Portkey’s guardrails ecosystem. This integration supports four powerful Azure services:

Azure Content Safety

A comprehensive content moderation service that detects harmful content including hate speech, violence, sexual content, and self-harm references in text.

Azure PII Detection

Advanced detection of personally identifiable information (PII) and protected health information (PHI) to safeguard sensitive data.

Azure Shield Prompt

Detects jailbreak and prompt injection attacks using Azure AI Content Safety Prompt Shields API.

Azure Protected Material

Detects known protected/copyrighted text content in LLM outputs to help ensure intellectual property compliance.

Setting Up Azure Guardrails

Follow these steps to integrate Azure’s content moderation services with Portkey:

1. Configure Azure Authentication

Navigate to the Integrations page under Settings to set up your Azure credentials. You can authenticate using three different methods:

• API Key - Uses a simple API key for authentication
• Entra (formerly Azure AD) - Uses Azure Active Directory authentication
• Managed - Uses managed identity authentication within Azure

Each authentication method requires specific credentials from your Azure account:

API Key

• Resource Name: Your Azure resource name
• API Key: Your Azure API key

Entra

• Resource Name: Your Azure resource name
• Client ID: Your Azure client ID
• Client Secret: Your client secret
• Tenant ID: Your Azure tenant ID

Managed

• Resource Name: Your Azure resource name
• Client ID: Your Azure client ID (for managed identity)

2. Create Azure Guardrail Checks

Once authentication is set up, you can add Azure guardrail checks to your Portkey workflow:

1. Navigate to the Guardrails page
2. Search for either Azure Content Safety or Azure PII Detection
3. Click Add to configure your chosen guardrail
4. Configure the specific settings for your guardrail
5. Save your configuration and create the guardrail

Guardrail Actions allow you to orchestrate your guardrails logic. You can learn more about them here

Azure Content Safety

Azure Content Safety analyzes text for harmful content across several categories.

Configuration Options

Parameter	Description	Values
Blocklist Names	Custom Blocklist names from your azure setup	blocklist-1, blocklist-2, blocklist-3
API Version	Azure Content Safety API version	Default: 2024-09-01
Severity	Minimum severity threshold for flagging content	2, 4, 6, or 8
Categories	Content categories to monitor	Hate, SelfHarm, Sexual, Violence
Timeout	Maximum time in milliseconds for the check	Default: 5000

Using Blocklists

Blocklists allow you to define custom terms or patterns to be flagged. You’ll need to create Content Safety blocklists in your Azure account first, then reference them in the Blocklist Names field.

For more information on Azure Content Safety blocklists, visit the official documentation.

Azure PII Detection

Azure PII Detection identifies and can help protect personal and health-related information in your content.

Configuration Options

Parameter	Description	Values
Domain	The type of sensitive information to detect	none (both PII and PHI) or phi (only PHI)
API Version	Azure PII Detection API version	Default: 2024-11-01
Model Version	Version of the detection model to use	Default: latest
Redact	Option to redact detected information	true or false
Timeout	Maximum time in milliseconds for the check	Default: 5000

Azure Shield Prompt

Azure Shield Prompt detects jailbreak and prompt injection attacks in your requests using Azure AI Content Safety Prompt Shields API. This guardrail analyzes system prompts and user messages to identify potential security threats before they reach your LLM.

Shield Prompt is designed to run on the input (before request) only. It cannot be used as an output guardrail.

Configuration Options

Parameter	Description	Values
API Version	Azure Content Safety API version	Default: 2024-09-01
Timeout	Maximum time in milliseconds for the check	Default: 5000

How It Works

Shield Prompt analyzes:

• System prompts: Checks if system messages contain potential attacks
• User documents: Analyzes user-provided content for injection attempts

When an attack is detected, the guardrail returns a failed verdict, allowing you to block or handle the malicious request appropriately.

For more information on Azure Prompt Shields, visit the official documentation.

Azure Protected Material

Azure Protected Material detects known copyrighted or protected text content in LLM outputs using Azure AI Content Safety Protected Material Detection API. This helps ensure your AI applications don’t inadvertently generate content that may infringe on intellectual property rights.

Protected Material is designed to run on the output (after request) only. It cannot be used as an input guardrail.

Configuration Options

Parameter	Description	Values
API Version	Azure Content Safety API version	Default: 2024-09-01
Timeout	Maximum time in milliseconds for the check	Default: 5000

How It Works

Protected Material scans the LLM’s response text and checks it against Azure’s database of known protected content. If protected material is detected, the guardrail returns a failed verdict, allowing you to:

• Block the response from being returned to the user
• Log the incident for compliance review
• Take alternative action based on your policies

Use Cases

• Compliance with copyright and intellectual property laws
• Preventing reproduction of licensed content
• Enterprise content governance

For more information on Azure Protected Material Detection, visit the official documentation.

Add Guardrail ID to a Config and Make Your Request

• When you save a Guardrail, you’ll get an associated Guardrail ID - add this ID to the input_guardrails or output_guardrails params in your Portkey Config
• Create these Configs in Portkey UI, save them, and get an associated Config ID to attach to your requests. More here.

Here’s an example configuration:

{
  "input_guardrails": ["guardrails-id-xxx", "guardrails-id-yyy"],
  "output_guardrails": ["guardrails-id-xxx", "guardrails-id-yyy"]
}

NodeJS

const portkey = new Portkey({
    apiKey: "PORTKEY_API_KEY",
    config: "pc-***" // Supports a string config id or a config object
});

Python

portkey = Portkey(
    api_key="PORTKEY_API_KEY",
    config="pc-***" # Supports a string config id or a config object
)

OpenAI NodeJS

const openai = new OpenAI({
  apiKey: 'OPENAI_API_KEY',
  baseURL: PORTKEY_GATEWAY_URL,
  defaultHeaders: createHeaders({
    apiKey: "PORTKEY_API_KEY",
    config: "CONFIG_ID"
  })
});

OpenAI Python

client = OpenAI(
    api_key="OPENAI_API_KEY", # defaults to os.environ.get("OPENAI_API_KEY")
    base_url=PORTKEY_GATEWAY_URL,
    default_headers=createHeaders(
        provider="openai",
        api_key="PORTKEY_API_KEY", # defaults to os.environ.get("PORTKEY_API_KEY")
        config="CONFIG_ID"
    )
)

cURL

curl https://api.portkey.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "x-portkey-api-key: $PORTKEY_API_KEY" \
  -H "x-portkey-config: $CONFIG_ID" \
  -d '{
    "model": "gpt-3.5-turbo",
    "messages": [{
        "role": "user",
        "content": "Hello!"
      }]
  }'

For more, refer to the Config documentation.

Monitoring and Logs

All guardrail actions and verdicts are visible in your Portkey logs, allowing you to:

• Track which content has been flagged
• See guardrail verdicts and actions
• Monitor the performance of your content moderation pipeline

Using Azure Guardrails - Scenarios

After setting up your guardrails, there are different ways to use them depending on your security requirements:

Detect and Monitor Only

To simply detect but not block content:

• Configure your guardrail actions without enabling “Deny”
• Monitor the guardrail results in your Portkey logs
• If any issues are detected, the response will include a hook_results object with details

Redact PII Automatically

To automatically remove sensitive information:

• Enable the Redact option for Azure PII Detection
• When PII is detected, it will be automatically redacted and replaced with standardized identifiers
• The response will include a transformed flag set to true in the results

Block Harmful Content

To completely block requests that violate your policies:

• Enable the Deny option in the guardrails action tab
• If harmful content is detected, the request will fail with an appropriate status code
• You can customize denial messages to provide guidance to users

---

Need Support?

If you encounter any issues with Azure Guardrails, please reach out to our support team through the Portkey community forum.