Azure API Management + Prisma AIRS Integration

The policy fragment contains the C# expressions that extract content, build the AIRS payload, send the scan request, and evaluate the verdict.

1. In the APIM instance, navigate to APIs → Policy fragments.
2. Click + Create.
3. Set Fragment ID to panw-airs-scan.
4. Paste the contents of the prisma-airs-policy-fragment-v1/panw-airs-scan file into the fragment editor.
5. Click Create.

The fragment behavior is controlled by context variables set before the <include-fragment> directive. These are set in the API policy XML, not in the fragment itself.

Add the AIRS fragment to the inbound policy section of the LLM API. This scans prompts before forwarding to the backend.

1. In the APIM instance, navigate to APIs → select the LLM API.
2. Click All operations (or a specific operation).
3. Open the Inbound processing policy editor.
4. Add the following after <base /> and any <set-backend-service> directive:

   <!-- Optional: Custom error messages -->
   <set-variable name="airsDescriptions" value="@{
       return new JObject(
           new JProperty("url_cats", "There are unacceptable URLs."),
           new JProperty("dlp", "There is sensitive data."),
           new JProperty("injection", "This contains content that is interpreted as trying to do something malicious."),
           new JProperty("toxic_content", "This violates our content policy."),
           new JProperty("malicious_code", "Malicious code detected."),
           new JProperty("topic_violation", "This is off topic."),
           new JProperty("db_security", "Unacceptable SQL commands detected."),
           new JProperty("ungrounded", "Content is not grounded.")
       );
   }" />
   
   <!-- Optional: Fail-open if AIRS is unreachable -->
   <set-variable name="FailOpen" value="false" />
   
   <!-- Recommended: Set profile name -->
   <set-variable name="currentProfile" value="YOUR_PROFILE_NAME" />
   
   <!-- Optional: Tool scanning configuration -->
   <set-variable name="scanTools" value="true" />
   <set-variable name="toolProfile" value="YOUR_TOOL_PROFILE" />
   
   <!-- Mandatory: Scan type and fragment include -->
   <set-variable name="ScanType" value="prompt" />
   <include-fragment fragment-id="panw-airs-scan" />

   Copy
5. Click Save.

Add the AIRS fragment to the outbound policy section. This scans LLM responses before returning to the client.

1. Open the Outbound processing policy editor for the same API.
2. Add the following after <base />:

   <!-- Optional: Different profile for response scanning -->
   <set-variable name="currentProfile" value="YOUR_RESPONSE_PROFILE" />
   
   <!-- Mandatory -->
   <set-variable name="ScanType" value="response" />
   <include-fragment fragment-id="panw-airs-scan" />

   Copy
3. Click Save.