Defend
ReferenceModules

Toxicity output

Detect toxic content in model output.

View on GitHub

YAML key: toxicity_output
Direction: output

Detect toxic content in model output.

Configuration

Prop

Type

Example

defend.config.yaml (fragment)
guards:
  output:
    enabled: true
    provider: claude
    modules:
      - toxicity_output:
          categories: []
          severity: "standard"

Provider usage

Configure under guards.output.modules with output guarding enabled and provider claude or openai.

See the modules overview.

Toxicity

Detect toxic or abusive content in user input.

Sensitive topics

Restrict sensitive political, medical, or other configured topics.

On this page

ConfigurationExampleProvider usage