HalluVerse25 introduces a comprehensive multilingual dataset specifically designed to identify and evaluate hallucinations in Large Language Models across different languages.

• Captures fine-grained hallucinations that many existing benchmarks miss
• Enables cross-lingual evaluation of LLM factual reliability
• Provides a framework for detecting entity-level, relation-level, and sentence-level hallucinations
• Supports security improvements by reducing misinformation risks

For security professionals, this research addresses a critical vulnerability in AI systems: the tendency to generate convincing but non-factual content that could lead to misinformation propagation or compromise decision-making in sensitive contexts.

HalluVerse25: Fine-grained Multilingual Benchmark Dataset for LLM Hallucinations