Preserving Professional Skepticism in the Age of Algorithmic Auditing: A Conceptual Framework for AI Adoption in Accounting Practice

Abstract

Generative artificial intelligence (AI) has moved beyond the transactional automation that characterized earlier accounting technologies such as robotic process automation and rule-based expert systems. Large language models and machine-learning-based analytics can now draft audit memoranda, flag anomalies in real time, and generate narrative explanations that resemble professional judgment rather than mere calculation. This shift raises a question that the accounting literature has only begun to address systematically: at what point does AI-generated output stop assisting professional judgment and start substituting for it, and what happens to auditors' professional skepticism along the way? Drawing on recent systematic reviews and empirical studies of AI in accounting and auditing, this paper develops a conceptual framework — the Skepticism-Preserving AI Adoption (SPAA) framework — that classifies accounting and audit tasks into three tiers according to the degree of contextual and ethical judgment they require, and links each tier to a specific set of human-control mechanisms: task decomposition, explainability thresholds, mandatory disconfirmation protocols, and accountability anchoring. The paper argues that automation bias, rather than technical unreliability, is the primary risk of AI adoption in high-judgment accounting tasks, and that professional skepticism must be deliberately engineered into workflows rather than assumed to survive automation by default. Implications for audit practice, standard-setting, and accounting education are discussed, along with testable propositions to guide future empirical research.