Civil liability and risk allocation in the use of generative artificial intelligence: a comparative analytical study

This study addresses the legal vacuum surrounding the allocation of civil liability for damages resulting from generative artificial intelligence (AI) outputs in creative and academic authorship, particularly given the “Black Box Problem” that hinders proving the “Harmful Act” (fault) [Thomson Reuters Enterprise Centre GmbH v Ross Intelligence Inc No. 1:20-cv-613-SB (D. Del. Feb. 11, 2025), appeal docketed (3d Cir. June 2025)]. While partial frameworks exist, such as Jordan’s 2022 National AI Code of Ethics, they remain insufficient as they are non-binding soft law lacking explicit civil enforcement mechanisms. Employing a research design that integrates doctrinal analysis and comparative methodology, the study evaluates traditional liability doctrines in Jordanian and Iraqi legislation against global regulatory developments, such as the EU AI Act, U.S. Executive Order 14,110 (2023), and the subsequent 2025 Executive Order 14,365. The primary contribution is the proposition of a structured “Liability Allocation Model,” empirically validated against simulated 2026 judicial scenarios, that distributes legal burdens based on functional control capacity and inherent risks. To resolve algorithmic opacity, the model enforces concrete transparency mechanisms, specifically “Mandatory Audit Logs,” enabling judicial reverse-engineering of outputs. The study concludes with the necessity of adopting a hybrid legislative approach combining fault-based liability for authors (users) and strict objective liability for developers of high-risk generative models, alongside providing binding technical disclosure mechanisms to ensure intellectual property protection and professional integrity (Ibid.).