Abstract
This chapter examines epistemic trust, AI-generated evidence, and legal liability in judicial proceedings. The study analyses the legal nature of AI-based evidence, identifying key characteristics — algorithmic generation, data dependency, and black box opacity — which challenge traditional evidentiary standards of relevance, admissibility, and verifiability. Drawing on comparative analysis of Wisconsin v. Loomis (2016), the Dutch SyRI judgment (2020), and the Toeslagenaffaire scandal, the chapter shows that algorithmic errors are structurally diverse, systemically discriminatory, and resistant to detection through standard legal mechanisms. Legal liability is explored across multiple actors — developers, operators, and judicial users — while critically assessing traditional fault-based models. The chapter concludes that effective regulation requires risk-based approaches combining algorithmic transparency, mandatory human oversight, and procedural safeguards — as reflected in the EU AI Act 2024 — to ensure AI in justice complies with the rule of law and human rights.

This publication has 18 references indexed in Scilit: