Legal Hallucinations in Arabic Legal NLP: An Empirical Study Toward a Diagnostic and Mitigation Framework

Abstract

Legal hallucination in Arabic large language models (LLMs) represents a growing challenge for the integrity of AI-assisted legal work. This study examines the structural relationship between the deficiencies of Arabic Legal Natural Language Processing systems and the emergence of legal hallucinations, with a particular focus on the Egyptian legal context. The research adopts a qualitative doctrinal methodology combined with an empirical evaluation testing two leading LLMs—ChatGPT and Gemini—using Egyptian statutory and judicial materials to observe and classify hallucination patterns in practice. The empirical analysis identified recurrent hallucination patterns within the exploratory sample: ChatGPT favored contextual errors, whereas Gemini produced more citation hallucinations. As demonstrated through empirical tests, the study offers a preliminary taxonomy of hallucinations within Arabic legal contexts, namely, citation hallucination, contextual hallucination, temporal hallucination, doctrinal hallucination, and hierarchical reasoning failure. The study makes an important contribution to the domain of Arabic Legal AI through the introduction of an empirical framework of analysis comprising three components.     

Received: 12 June 2026 | Revised: 13 August 2026 | Accepted: 22 September 2026

Conflicts of Interest

The author declares that he has no conflicts of interest to this work.

Data Availability Statement

The datasets generated and analyzed during the current study are contained within the article and its supplementary Appendix A. No additional datasets were generated beyond those reported in the manuscript.

Author Contribution Statement

Hossam Mostafa Fawzy: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing – original draft, Writing – review & editing, Visualization.