Answers are grounded in this paper and cite their sources.
As public administrations embed artificial intelligence (AI) in mandatory services such as income taxation, discretion shifts from case-level judgment to system-level design choices. Privacy then concerns not only which data are held but how citizens are judged, as algorithms infer sensitive attributes and categorize individuals. Because citizens can neither opt out of these services nor audit the underlying logic, privacy frameworks built on voluntary use and personal control are inadequate to explain how citizens’ concerns arise or how regulation addresses them. This study draws on a dyadic model in which legitimacy requires shared expectations, perceived conformity to them, and assent. Two AI misuse concerns, reduced judgment and algorithm bias, are conceptualized as perceived violations of the procedural justice expectations: correctability and bias suppression. The resulting model treats perceived nonconformity as an antecedent of assent and regulation as a factor shaping these concerns. It is estimated using partial least squares structural equation modeling on a scenario-based online survey of 312 employed German residents who evaluated AI-based risk classification of tax returns. Perceived effectiveness of the EU’s General Data Protection Regulation is positively associated with trust in the tax administration but not directly with either concern. Instead, trust fully mediates its associations with the two concerns. Both concerns, in turn, are associated with a less favorable attitude toward the AI procedure. Where citizens cannot exit, regulatory assurance thus appears to work through institutional trust, so administrations should look beyond compliance and embed visible procedural safeguards that build this trust.
No takes yet. Share an insight, caveat, or question.
Wojke et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: