A model trained to categorize or label text into predefined categories, such as whether content is safe or violates policies.