Slipping Through the Cracks, the Carve-outs for AI Tax Enforcement Systems in the EU AI Act

Tax administrations are among State organs who leverage AI-technology the most. AI systems used by tax administrations have already led to scandals and seminal jurisprudence such as SyRI, eKasa or SS SIA. The most striking example is the Dutch toeslagenaffaire where using an AI model, the tax administration discriminated and profiled taxpayers based on their ethnicity, causing irreparable harms. Growing awareness over the proliferation of AI triggered the Commission to submit the proposal for the European Union Artificial Intelligence Regulation (EU AI Act) as a regulatory response to control these externalities. Yet, confusion remains around the treatment of AI systems leveraged by tax administrations in the AI Act. Without a category of their own, it is unclear whether AI systems used by tax administrations qualify as high-risk systems or not. This uncertainty in the current version of the proposal raises the following question: “based on a teleological interpretation of the draft proposal for the EU AI Act, should AI systems used by tax administrations be regarded as high-risk systems?” This question is addressed in two parts. Section 1 presents the state of use of AI systems by tax administrations in the entire EU and the typology of functions performed by these systems. Section 2 examines whether, under the current text of the proposal for the EU AI Act and the different positions of EU institutions, these systems should be regarded as high-risk systems.