AI Moderation: Inconsistencies in Hate Speech Detection Across LLM-Based Systems