Artificial intelligence, as an emerging phenomenon, has created significant challenges in the field of civil liability despite its considerable benefits. AI systems, due to their complexity and operational autonomy, have cast doubt upon the traditional notions of causation and fault. From the perspective of corrective justice, the development of liability rules proportionate to the nature of such systems is essential, as their autonomous behavior reduces predictability and expands the scope of potential harm. Nevertheless, even in cases of fully autonomous operation, the role of human actors in the chain of development, production, and management cannot be disregarded. Accordingly, the rules governing civil liability should be designed in a manner that creates effective incentives for reducing the social costs arising from AI-related accidents. Adopting an economic efficiency approach, this article seeks to answer the question of which liability theory is more efficient for each of these two groups. The findings indicate that strict liability for developers and fault-based theory for managers can, respectively, contribute to achieving economic efficiency. This combined model, while creating preventive incentives, can establish a balance between protecting injured parties and encouraging innovation.