Bilingual Legal Citation Fabrication Benchmark: Saudi PDPL and GDPR (Arabic–English)

A bilingual benchmark of 120 questions (60 per law, each in parallel Arabic and English) that probes whether large language models cite the correct article of a data-protection law, abstain when no article applies, or fabricate a citation. It covers the EU General Data Protection Regulation (GDPR) and the Saudi Personal Data Protection Law (PDPL, as amended by Royal Decree M/148), with direct article-retrieval questions, true- and false-premise verification questions, and unanswerable trap questions, including a repealed-article probe and deadlines that exist only in the PDPL Implementing Regulations. The record includes the question files with gold answers and trap rationales, a bilingual index of all 43 articles of the amended PDPL, the evaluation script with the exact system prompts, raw outputs of three freely accessible models (Gemini 2.5 Flash, GPT-OSS-120B, Nemotron-3-Super-120B; 645 responses), and the scored results and summaries. See README.md for the dataset card, settings, scoring taxonomy, and instructions for reproducing the results. This dataset accompanies: N. S. Alrajeh, "Do LLMs Fabricate Legal Citations? A Bilingual Benchmark on Saudi Data Protection Law and the GDPR," arXiv:2607.11127, 2026.