Llama Guard 4 is a specialist, not a generalist — it exists solely to classify whether content is safe or unsafe, not to answer questions or generate text. It processes both text and images, making it multimodal in its safety assessments. Think of it as a focused auditor: narrow in purpose, deliberate in judgment, and designed to slot into pipelines rather than serve end users directly.