This dataset contains 4,000 sanitized prompts designed for research on prompt injection vulnerabilities in Large Language Models (LLMs).It introduces and supports evaluation of a novel attack class called Narrative-Embedded Prompt Injection (NEPI), where adversarial intent is embedded inside coherent fictional narratives, dialogues, or persona-driven roleplay prompts.
Unlike traditional… See the full description on the dataset page: https://huggingface.co/datasets/Vaibhav-GOAT/nepi-prompts-dataset.
