Dataset Catalog

BeaverTails is a collection of AI‑safety‑focused datasets containing a series of human‑annotated question‑answer pairs, each labeled with a corresponding harm category. The dataset covers 14 harm categories such as animal abuse, child abuse, discrimination, hate speech, etc. It is intended for research, especially for creating safer, less‑harmful AI systems. The dataset includes multiple splits: 330k_train, 330k_test, 30k_train and 30k_test.