Answers are grounded in this paper and cite their sources.
This article examines legal frameworks for algorithmic transparency in the EU and US, suggesting improvements for accountability and auditability.
This article examines algorithmic opacity as a legal–technical problem that undermines the enforceability of transparency and accountability obligations in automated decision-making systems across the European Union (EU) and the USA. From a comparative legal perspective, it argues that opacity is not merely a matter of technical complexity but an institutional condition that prevents verifiable accountability, hinders the traceability of automated decisions, and weakens the practical protection of equality, non-discrimination, privacy, and due process. The analysis shows that the EU, through Regulation (EU) 2024/1689, adopts a preventive and risk-based approach that translates transparency and intelligibility into ex ante documentation, traceability, and impact assessment duties, whereas the US framework remains fragmented, relying on sectoral or state-level rules and predominantly ex post enforcement. Landmark cases (SyRI, iBorderCtrl, COMPAS, SafeRent) illustrate how the absence of auditable technical evidence produces disproportionate effects on vulnerable groups and limits effective contestability. The article further highlights technological and extraterritorial asymmetries, as many systems deployed in the EU are designed or operated by US-based providers, constraining access to the technical evidence required for legal scrutiny. In response, it proposes an operational standard of ‘sufficient intelligibility’, grounded in a minimum package of technical evidence—model and data memory, inference records, and subgroup performance metrics—combined with a risk-proportionate verification scheme (ex ante/ex post). This framework operationalizes transparency as a verifiable legal obligation, enabling courts, regulators, and internal control bodies to assess, audit, and enforce accountability in practice.
No takes yet. Share an insight, caveat, or question.
García-Llorente et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: