Privasis is a large-scale synthetic dataset consisting of diverse text records—such as medical and financial records, legal documents, emails, and messages—containing rich, privacy-sensitive information. Each record includes synthetic profile details, surrounding social context, and annotations of privacy-related content. All data are fully generated using GPT-OSS-120B, supplemented with the personas from the public Nemotron-Personas-USA dataset… See the full description on the dataset page:
https://huggingface.co/datasets/nvidia/Privasis-USA.
