Explainable Artificial Intelligence and Data Governance in U.S. Financial Reporting: A Scoping Review of Reporting Quality, Regulatory Accountability, and Decision Transparency

Abstract

Background: Artificial intelligence is increasingly used in financial reporting, auditing, disclosure preparation, internal-control monitoring and regulatory compliance in United States organisations. However, opaque model outputs, weak data governance and uncertain human accountability may limit reporting quality, auditability and decision transparency. Aim: This scoping review mapped evidence on how explainable artificial intelligence and data governance support financial reporting quality, regulatory accountability and decision transparency in United States organisations. Methods: The review used a scoping design guided by the Population-Concept-Context framework and PRISMA-ScR principles. Evidence was identified from Scopus, IEEE Xplore and the U.S. Securities and Exchange Commission website. Eligible sources were English-language materials published between 2022 and 2026 that addressed artificial intelligence, explainability, data governance, model governance, auditability, internal control, disclosure or accountability in accounting, auditing or reporting contexts. Data were charted on AI/XAI applications, governance mechanisms, reporting-quality domains, regulatory-accountability links and evidence gaps. Results: Nineteen sources were included, comprising peer-reviewed articles, working papers, preprints, professional guidance and regulatory documents. Applications included generative AI, machine learning, decision trees, random forests, fraud detection, anomaly detection, disclosure analytics and audit decision aids. Governance mechanisms included explainability tools, confidence scores, human overrides, validation datasets, cybersecurity controls, audit trails, board oversight and prompt/output documentation. The evidence linked AI to reporting timeliness, ledger granularity, disclosure analysis, audit support, internal-control monitoring and misstatement-risk reduction. Nevertheless, data lineage, vendor governance, retention rules, approval workflows and formal AI assurance remained underdeveloped. Conclusion: AI-supported reporting is most defensible when data, models, controls and human judgements are explainable, traceable, reviewable and accountable.