Policy: Azure > AI Foundry > RAI Policy > Content Filters

Define the Content Filters configuration required for Azure > AI Foundry > RAI Policy.

A RAI policy is a list of content filters. Each filter has a name (for example Jailbreak), a source (Prompt for the input filter or Completion for the output filter), whether it is enabled, whether it blocks, and for the four harm categories a severity threshold. The sub-policies under Content Filters > Input Filter and Content Filters > Output Filter let you require a state for each filter; Content Filters > Mode covers the policy's streaming mode. Every sub-policy defaults to Skip, so only the filters you set are checked.

When set to Check: Configured the control goes to alarm if any configured sub-policy is not met, and its details list every filter with the current and expected value. When set to Enforce: Configured the control updates the RAI policy in place, adding or correcting only those filters and leaving every other setting untouched. Skip excludes the RAI policy from this check.

System-managed RAI policies (Microsoft.Default, Microsoft.DefaultV2) cannot be modified in Azure; the control skips them.