Safety and Fairness for Content Moderation in Generative Models