A Databricks GenAI application is being designed for this workflow: a source corpus contains toxic forum posts that are useful for troubleshooting but unsafe to quote for a safety incident summarizer during a production readiness review. The assistant should answer safely without reproducing slurs. The design question is problematic text for a safety incident summarizer. What should the engineer recommend?
