This study aims to systematically investigate the evolution of theories, models, and techniques in the field of accounting fraud detection from 2000 to 2025. As fraudulent financial reporting continues to undermine the credibility of accounting information and destabilize global economic systems, there is an urgent need to critically evaluate existing detection frameworks and technologies. Employing a qualitative research design based on a systematic literature review methodology, this study analyzed 150 peer-reviewed academic sources across multidisciplinary databases using thematic synthesis. The analysis was guided by a structured protocol that included defined inclusion and exclusion criteria, thematic coding, and data triangulation using NVivo software. The findings reveal a significant theoretical progression from foundational models like the Fraud Triangle to more complex and integrative frameworks, including the Fraud Diamond, the MICE model, and behavioral theories such as the “Dark Triad.” In parallel, empirical evidence highlights the growing dominance of machine learning and hybrid analytical models over traditional statistical techniques in fraud detection. Emerging technologies such as blockchain, big data analytics, and natural language processing are found to be instrumental in enhancing real-time detection capabilities, though challenges remain in interpretability, ethical governance, and data security. The study also identifies critical research gaps, including the need for cross-cultural validation, longitudinal analysis, and interdisciplinary collaboration. These findings contribute to both academic discourse and managerial practice by offering a comprehensive and forward-looking synthesis, serving as a foundational reference for future innovation in accounting fraud prevention.