Skip to content

Azure Content Safety Content Moderation

The Azure Content Safety guardrail integrates with the Azure Content Safety API to filter harmful content in requests and LLM-generated responses. It checks for four categories — hate speech, sexual content, self-harm, and violence — each with a configurable severity threshold.

Prerequisites

This guardrail requires an active Azure subscription with the Azure Content Safety service enabled.

  1. In the Azure Portal, search for Content Safety and create a new resource.
  2. After the resource is created, open it and go to Keys and Endpoint.
  3. Copy the Endpoint URL and one of the API Keys.
  4. Add them to your gateway's config.toml file:
azurecontentsafety_endpoint = "https://your-resource.cognitiveservices.azure.com"
azurecontentsafety_key = "your-api-key"

Configuration Parameters

Advanced Settings

Content moderation settings are configured independently for the request and response phases. Both sections have the same parameters:

Parameter Default Description
JSON Path — JSONPath to extract the content to evaluate. If empty, the entire payload is evaluated as plain text.
Passthrough On Error false When true, requests continue if the Azure API call fails. When false, an error is returned on API failure.
Show Assessment false When true, the intervention response includes detailed category scores.
Hate Severity Threshold 4 Severity threshold for hate speech (0–7). Use -1 to disable. Content at or above the threshold is blocked.
Sexual Severity Threshold 5 Severity threshold for sexual content (0–7). Use -1 to disable.
Self Harm Severity Threshold 3 Severity threshold for self-harm content (0–7). Use -1 to disable.
Violence Severity Threshold 4 Severity threshold for violence (0–7). Use -1 to disable.

Severity Scale

Score Meaning
0–1 Safe or negligible
2–3 Low severity
4–5 Medium severity
6–7 High severity

Add This Policy

  1. Navigate to AI Workspace > LLM Providers or LLM Proxies.
  2. Click on the provider or proxy name.
  3. Go to the Guardrails tab.
  4. Click + Add Guardrail and select Azure Content Safety Content Moderation from the sidebar.
  5. Expand Advanced Settings to configure severity thresholds and error handling separately for request and response phases.
  6. Click Add (for providers) or Submit (for proxies).
  7. Deploy the provider or proxy to apply the changes.

Behavior

  • When content meets or exceeds a configured severity threshold, the gateway returns 422 Unprocessable Entity with a guardrail intervention response.
  • Request and response phases are evaluated independently using their respective settings.
  • If Show Assessment is enabled, the response includes the category scores that triggered the block.