Artificial Intelligence in Tax Enforcement under the EU AI Act: The Relevance of European Convention on Human Rights Ne Bis in Idem Jurisprudence for High-Risk Classification
Christina Dimitropoulou · World Tax Journal · 2026
Artificial intelligence (AI) systems are increasingly deployed by EU tax administrations to detect non-compliance, generate risk scores and assist in imposing administrative penalties. The EU AI Act (Regulation (EU) 2024/1689) classifies the most consequential systems as “high risk”, yet Annex III excludes tax administration AI, and Recital 59 exonerates such systems unless they serve a criminal investigative purpose. This article argues that this treatment is constitutionally inadequate, producing a regulatory gap irreconcilable with the European Union’s fundamental rights framework. First, the AI Act is subject to the Charter of Fundamental Rights, whose article 52(3) makes ECHR case law the minimum standard for corresponding Charter rights. Second, drawing on the Engel doctrine and Åkerberg Fransson (C-617/10), the article shows that AI-shaped administrative tax penalties already carry a punitive character, triggering criminal protections, while De Legé v. the Netherlands (App. No. 58342/15) shows that the guarantees under article 6 of the ECHR engage when evidence is collected, thereby undermining the Act’s administrative/criminal distinction. Third, the ne bis in idem “sufficiently close connection” test from A and B v. Norway is often compromised by opaque algorithmic reasoning inaccessible to taxpayers and courts. The article proposes amending Annex III and article 7 to incorporate the Engel test as a proxy for fundamental rights equivalence.