Evil Geniuses: Delving into the Safety of LLM-based Agents