Generative AI in Accounting Policies and Estimates: Safeguarding Professional Judgment and Addressing Managerial Bias

Abstract

Generative artificial intelligence (GenAI) is increasingly used in corporate reporting, where accounting policies and estimates require professional judgment. The central risk examined is not automated error but automated advocacy: a fluent, standard-referenced justification for a preferred treatment can now be obtained in seconds, making biased judgment a governance problem before it is an accounting problem. Using theory adaptation, this conceptual study integrates research on professional judgment, motivated reasoning, and automation reliance. It identifies five bias-amplification mechanisms: anchoring, automation bias, outcome-directed prompting, unverified justification, and diffusion of responsibility. It develops TRACE, a proposed framework comprising task boundary, regulatory grounding, alternative assessment, challenge and contradiction, and evidence trail. Its intended operation is illustrated using four constructed IFRS cases (IAS 8/IAS 16, IAS 36, IAS 37, IFRS 15) and a boundary-condition variant in which TRACE does not resolve the disagreement. The AI proposals are stylised illustrations, not model outputs or empirical data. GenAI appears most defensible when used to generate alternatives and challenge a preferred treatment, and least defensible when asked to select or justify a conclusion. TRACE may make AI-assisted judgments more attributable and contestable, but it has not been empirically validated and cannot guarantee unbiased outcomes.