The liability vacuum: Why existing legal frameworks cannot govern artificial general intelligence and what must be done

Abstract

Existing artificial intelligence (AI) liability scholarship focuses largely on narrow, task-specific systems. This article asks how existing liability frameworks would function – or fail – when confronted with artificial general intelligence (AGI). It examines four liability regimes – strict product liability, fault-based negligence, vicarious liability and strict liability for hazardous activities – across five jurisdictions (the European Union, United States, United Kingdom, Singapore and China), concentrating its comparative analysis on product liability, the regime most affected by digital-era reform, while situating the other three within the broader structural critique that follows. These regimes embed assumptions about predictability, human control, causal traceability and task-specificity that AGI would structurally violate. AGI challenges liability law not merely evidentially but structurally, destabilizing the core legal concepts of defectiveness, foreseeability, proximate causation and attribution of agency on which existing regimes depend. This discussion reflects the current regulatory context: the withdrawal of the EU AI Liability Directive proposal and the adoption of the revised Product Liability Directive (Directive [EU] 2024/2853). Drawing on nuclear and environmental liability analogies, the article proposes an anticipatory liability architecture: channeled strict liability with mandatory capitalization, capability-calibrated duties of care and an international compensation fund.