This research introduces ELAB, the first extensive framework for evaluating Persian language models across crucial ethical dimensions.

• Creates three types of Persian benchmarks: translated, synthetically generated, and naturally collected data
• Develops SafeBench-fa to test models specifically for Persian cultural and linguistic contexts
• Addresses critical gaps in LLM evaluation frameworks for non-English languages
• Establishes security and safety benchmarks to prevent harmful content generation

This work matters for security professionals because it provides standardized tools to evaluate AI safety across linguistic and cultural boundaries, essential for global AI deployment without safety compromises.

ELAB: Extensive LLM Alignment Benchmark in Persian Language