Artificial intelligence (AI) increasingly influences contemporary criminal justice. However, current European instruments, most notably the EU Artificial Intelligence Act (2024) and the Council of Europe AI Convention (2024), leave significant procedural and substantive gaps. This article argues that criminal justice requires a specific lex specialis capable of addressing issues that broader AI governance does not resolve, particularly evidentiary standards, due process, and liability allocation. The article presents three core claims. First, it proposes a Glass-Box Standard for AI-mediated evidence and decisions. Under this standard, contestability, auditability, and reproducibility are essential conditions for admissibility or, at least, for assigning appropriate probative weight. Second, it outlines a tri-layer liability model that distinguishes the responsibilities of developers, deployers, and decision-makers, and considers the absence of a CriminalJustice AI Impact Assessment (CJ-AIA) as negligence per se. Third, drawing on comparative insights from the United States (State v. Loomis), the United Kingdom (Bridges), and the Netherlands (SyRI), the article demonstrates how persistent problems of transparency, proportionality, and equality remain inadequately addressed by existing regulatory frameworks. Based on this analysis, the article offers a reform blueprint that includes mandatory CJ-AIA; strengthened duties of disclosure, logging, and version control; an algorithmic impeachment right enabling practical defence challenges; and calibrated exclusionary rules for AI-tainted evidence. These measures seek to ensure that AI can support criminal justice without undermining fundamental rights. The overarching aim is to reconcile technological innovation with the rule of law. AI can assist criminal proceedings, but only when integrated within transparent, accountable, and human-centred legal frameworks capable of safeguarding liberty, equality, and procedural fairness.
Towards a Lex Specialis for AI in Criminal Justice: Evidence, Liability, and fundamental rights in the age of Algorithmic decision-making
vuosi
Writing – Original Draft Preparation
2025
Abstract
Artificial intelligence (AI) increasingly influences contemporary criminal justice. However, current European instruments, most notably the EU Artificial Intelligence Act (2024) and the Council of Europe AI Convention (2024), leave significant procedural and substantive gaps. This article argues that criminal justice requires a specific lex specialis capable of addressing issues that broader AI governance does not resolve, particularly evidentiary standards, due process, and liability allocation. The article presents three core claims. First, it proposes a Glass-Box Standard for AI-mediated evidence and decisions. Under this standard, contestability, auditability, and reproducibility are essential conditions for admissibility or, at least, for assigning appropriate probative weight. Second, it outlines a tri-layer liability model that distinguishes the responsibilities of developers, deployers, and decision-makers, and considers the absence of a CriminalJustice AI Impact Assessment (CJ-AIA) as negligence per se. Third, drawing on comparative insights from the United States (State v. Loomis), the United Kingdom (Bridges), and the Netherlands (SyRI), the article demonstrates how persistent problems of transparency, proportionality, and equality remain inadequately addressed by existing regulatory frameworks. Based on this analysis, the article offers a reform blueprint that includes mandatory CJ-AIA; strengthened duties of disclosure, logging, and version control; an algorithmic impeachment right enabling practical defence challenges; and calibrated exclusionary rules for AI-tainted evidence. These measures seek to ensure that AI can support criminal justice without undermining fundamental rights. The overarching aim is to reconcile technological innovation with the rule of law. AI can assist criminal proceedings, but only when integrated within transparent, accountable, and human-centred legal frameworks capable of safeguarding liberty, equality, and procedural fairness.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


