BeaverTails is a collection of datasets designed to facilitate research on safety alignment in large language models (LLMs).