Artificial Intelligence in Tax Administration: A Conceptual Examination of Automation, Compliance, Revenue Generation and Challenges Facing Modern Tax Systems

Abstract

Tax administration occupies a distinctive position in the broader literature on artificial intelligence (AI) in public administration: it is simultaneously among the sectors where AI adoption is most advanced, with major revenue authorities operating dozens of production AI models, and among the sectors where AI's failures have produced some of the most consequential legal reversals of automated government decision-making to date. This paper offers a conceptual examination of AI's role in tax administration that treats both facts as central rather than treating the second as an unfortunate footnote to the first. Integrating the economic deterrence model of tax compliance, the slippery-slope trust-and-power framework, and agency theory with recent evidence on AI-enabled automation, predictive risk profiling, and taxpayer-facing natural-language systems, the paper argues that AI strengthens tax administration's capacity to detect non-compliance and streamline routine processing considerably more reliably than it strengthens the trust relationship on which voluntary compliance, and therefore sustainable revenue generation, actually depends. Drawing on documented failures of automated tax and welfare decision-making, including Australia's Robodebt scheme and the Netherlands' System Risk Indication program, the paper shows that AI-enabled compliance systems can simultaneously become more accurate in the aggregate and less legitimate in the particular, with direct consequences for the voluntary-compliance foundation the slippery-slope framework identifies as tax administration's most efficient long-run outcome. The paper develops an integrated model connecting automation, compliance, and revenue generation through a governance layer addressing explainability, oversight, and redress, and considers the implications of that model for tax administrations, taxpayers, and future research.