Benchmarks and Datasets for Large Language Model Security: Coded Corpus (v1.0)

Coded corpus of 325 primary studies from a systematic mapping study of publicly released benchmarks and datasets for large language model (LLM) security, covering the period 2022–2026 (up to 10 June 2026). Each record contains: bibliographic identifiers (BibKey, title, year, venue, arXiv/DOI), primary threat category (one of 11 categories T1–T11 following a CIA-triad prompt-attack taxonomy), secondary threat tags (multi-label; 158 tags across 110 studies), six cross-cutting facets (modality, language coverage, target system, domain, evaluation paradigm, CIA dimension), and a W3C PROV provenance-maturity level (P0–P4) indicating the completeness of the artifact release.The corpus is the data foundation for the paper: "Benchmarks and Datasets for Large Language Model Security: A Systematic Mapping Study" Supplementary files included: - llm-security-benchmarks-coded-corpus-v1.0.xlsx — the complete coded corpus (Coding Sheet, Lists vocabulary, Legend tabs) - segress_checklist.pdf — SEGRESS reporting checklist - prisma_checklist.pdf — PRISMA 2020 reporting checklist