Legal Liability for Artificial Intelligence-Assisted Nuclear Safety Decisions: A Comparative Assessment of Operator Responsibility, Regulatory Accountability, and Compensation for Nuclear Damage

Abstract

regulatory work through anomaly detection, predictive maintenance, document review, simulation, information retrieval, and decision support. The legal consequence is not that established responsibility disappears, but that decision-making becomes distributed across operators, regulators, software suppliers, data providers, and technical support organizations. This article examines how existing nuclear law should allocate responsibility when an AI-assisted decision contributes to unsafe operation, defective regulatory action, or nuclear damage. It uses doctrinal and comparative analysis of the Vienna Convention on Civil Liability for Nuclear Damage, selected national nuclear liability regimes, IAEA safety standards, emerging work in the nuclear sector on AI, and peer-reviewed scholarship on AI liability and nuclear law. Three legal questions are separated: operational responsibility for safety, public-law accountability for regulatory decisions, and civil compensation for nuclear damage. The analysis finds that AI should not displace the nuclear operator’s prime responsibility for safety. Under channeled nuclear liability regimes, victims ordinarily claim against the operator without proving fault, while claims against suppliers are restricted, and operator recourse is limited. Defective AI may therefore alter technical causation without necessarily altering the victim-facing defendant. This makes contractual governance important: operators should preserve validation duties, audit rights, access to technical records, cybersecurity obligations, and express recourse clauses in the procurement of safety-significant AI. Where a regulator uses AI, statutory powers should remain attributable to authorized human officials; AI should support, not autonomously exercise, licensing, inspection or enforcement authority. Nigeria is a party to the 1963 Vienna Convention but not the 1997 Protocol or the Convention on Supplementary Compensation for Nuclear Damage. The article argues that AI provides an additional reason for Nigeria to review both nuclear liability modernization and nuclear digital-governance rules. A workable framework should preserve clear human and institutional responsibility while requiring traceability, validation, and reviewability across the AI decision chain.