Explore high-quality datasets for your AI and machine learning projects.
Aegis AI Content Safety Dataset 1.0 is an open‑source content safety dataset (CC‑BY‑4.0) that follows Nvidia's content safety taxonomy, covering 13 key risk categories. It includes approximately 11,000 human‑annotated interaction records between humans and LLMs, split into 10,798 training samples and 1,199 test samples. The data originate from Anthropic HH‑RLHF and Mistral‑7B‑v0.1, annotated by 12 annotators and 2 data quality assurance personnel. The dataset is intended for building content‑moderation safeguards and aligning LLMs to generate safe responses, but is not suitable for training dialogue agents. Its creation involved strict QA and annotator training to ensure diversity and accuracy.
