A University of Queensland study has shown Large Language Models (LLMs) used in AI content moderation may be prone to subtle ...