shanemhamilton/llm-prompt-guard-tuning-corpus

llm-prompt-guard tuning corpus Prompt-injection detection corpus used to tune the llm-prompt-guard pattern set. Two JSONL files: attacks.jsonl — 198 rows, label: 1. Injection payloads grouped by attack category (instruction override, role hijacking, jailbreak, unicode/homoglyph/tag-block smuggling, encoding bypass, and more). benign.jsonl — 1,310 rows, label: 0. Ordinary user input across seven domains, including phrasing that superficially resembles an attack ("ignore the… See the full description on the dataset page: https://huggingface.co/datasets/shanemhamilton/llm-prompt-guard-tuning-corpus.