The ongoing digitalization trend has not spared cantonal tax administrations. This article examines the increasing automation of tax assessment in light of the principle of legality, the duty to review and investigate, and the duty to provide justification.
Quick ReadThe ongoing digitalization trend has not spared cantonal tax administrations. This article examines the increasing automation of tax assessment in light of the principle of legality, the duty to review and investigate, and the duty to provide justification. The starting point is the observation that modern administrative procedures require increased efficiency and effectiveness, and that algorithmic systems—sometimes supplemented by AI—can make a significant contribution to this. At the same time, the question arises as to how these new technologies can be legally legitimized and embedded within existing legal frameworks. The principle of legality requires a statutory basis for the processing of personal data as well as for interventions in legal positions protected by fundamental rights. The authors demonstrate that this basis for conventional rule-based automation is usually derived from existing tax laws, whereas AI-supported systems that cannot be explained causally require a specific legal basis.
Particular emphasis is placed on the risk-based approach, which takes into account the varying intensity of interference and enables technology-neutral regulations. In practice, this means that automation projects require careful review with regard to admissibility, data categories, transparency, procedural rights, and quality management. The increasing automation of tax return reviews is deemed efficient and compatible with the tax authorities’ duty to review, provided that sufficient quality assurance measures are in place. By contrast, automation faces limitations in substantive examinations, as these require a case-by-case assessment and the exercise of discretion.
Regarding the duty to provide reasons, it is noted that at least standardized brief justifications are required whenever assessments deviate from the tax return. Conversely, data-driven AI systems without traceable* *decisions are not permissible unless a paradigm shift toward prospective and retrospective system controls is implemented, accompanied by corresponding constitutional or legislative amendments. Overall, the authors advocate for balanced regulation that harnesses the potential of automation without undermining the principles of the rule of law. A legal basis at the level of the Federal Tax Act (StHG) and the Federal Income Tax Act (DBG) or cantonal tax laws is deemed necessary to ensure both legitimacy and scope for innovation.
