Screen harmful or unwanted content from generative AI tasks.
Define what undesirable content the guardrails should detect.