Calibrating Algorithmic Accountability: Risk-Based Regulation, Enforcement Capacity and Deferral in European Artificial Intelligence Governance

Аңдатпа

Risk-based regulation has become the dominant paradigm for governing artificial intelligence, and Regulation (EU) 2024/1689 (the AI Act) is its most complete legislative expression. This article argues that the paradigm's principal weakness lies not in its substantive design but in the mismatch between the obligations it imposes and the institutional capacity available to enforce them. Drawing on doctrinal and comparative-institutional analysis of the AI Act's staged application, the unresolved position of the proposed AI Liability Directive, and parallel developments in the United States, the United Kingdom, China and the Council of Europe, the article identifies three structural pathologies. The first is a deferral dynamic: because risk-based regimes concentrate cost in ex ante conformity assessment, and that cost is borne before any harm materialises, compliance deadlines become politically renegotiable – a pressure already visible in the distance between the August 2026 date for Annex III high-risk obligations and the state of the harmonised standards on which conformity assessment depends. The second is a remedial asymmetry: with the AI Liability Directive stalled, public supervision has no matching private enforcement channel, so deterrence rests almost entirely on administrative capacity. The third is a capacity gradient: the transaction costs of risk-based regulation fall disproportionately on jurisdictions with thin supervisory institutions, which makes wholesale transplantation of the European model inadvisable for many emerging economies. The article proposes a capacity-indexed sequencing model in which transparency duties, procedural rights and sectoral supervision precede general-purpose conformity assessment.