Models and datasets used to improve the safety of LLMs, preventing harm in the presence of jailbreaking.