We need to better understand how LLMs address moral questions if we're to trust them with more important tasks.