The DynaGuardrail dataset is a comprehensive collection of safe and unsafe prompts designed to evaluate and improve guardrail systems for large language models (LLMs). The dataset comprises prompts across four critical safety domains: general safety, financial advice prohibition, tax advice prohibition, and prompt injection protection. Each instance is carefully labeled and validated by expert compliance officers and… See the full description on the dataset page: https://huggingface.co/datasets/dynamofl/DynaGuardrail.
