Artificial intelligence and earnings management: Governance strategies and model adoption

Abstract

Despite growing interest in using artificial intelligence (AI) to examine earnings management (EM), the literature remains fragmented across models and national contexts, offering limited comparative insight. This study maps the regional use of AI models in research on EM and related forms of accounting manipulation, including fraud detection and financial distress proxies. It is based on a systematic review of 21 peer-reviewed articles published between 2016 and 2025 in Scopus and Web of Science (WoS), following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Moher et al., 2009). The findings show that East Asian studies predominantly apply deep neural networks (DNN), deep belief networks (DBN), and hybrid models, often incorporating environmental, social, and governance (ESG)-related variables. West Asian research remains limited and mainly relies on natural language processing (NLP) of annual reports. North American studies primarily employ artificial neural networks and intelligent agents within fraud detection frameworks, while European research continues to use traditional indicators such as the Beneish M-score and Altman Z-score as empirical proxies. Overall, the study concludes that the effectiveness of AI-based approaches in addressing EM and related manipulative practices is institutionally contingent, highlighting the need for region-specific governance frameworks and cross-disciplinary collaboration.