DOI: 10.53136/97912218284673
Pages: 15-22
Publication date: September 2026
Publisher: Aracne
ABSTRACT This Article examines the application of the Portuguese regime of non-contractual State liability to damage caused by artificial-intelligence systems used by the Public Administration. It argues that liability should depend on the system’s degree of autonomy and human involvement: fault-based liability may apply to rule-based systems, whereas machine-learning systems call for a risk-based approach. Effective human oversight and a specific framework for State liability are therefore essential.
KEYWORDS Artificial intelligence - State liability - Public Administration - Human in the loop - Risk-based liability
TABLE OF CONTENTS 1. Introduction. – 2. Non-Contractual State Liability for Damages Arising from Public Action. – 3. Non-Contractual State Liability for Damages Arising from the Use of Artificial Intelligence in Public Administration Activity. – 4. Rule-Based Models, Linear Causality and Fault-Based Liability. – 5. Artificial Intelligence, Machine Learning Algorithms and Risk Liability. – 6. Decision-Making Support Models and Human in the Loop. – 7. Concluding Remarks.
