Ethical Safety Engineering in Production: A Conceptual Framework for Allocating Responsibility in AI-Controlled Production Systems

Abstract This paper addresses a significant and topical issue related to occupational safety in production systems using artificial intelligence. It presents an original conceptual framework enabling the systematic allocation of responsibility in such systems. This new approach takes into account the distributed nature of agency, algorithmic decision autonomy, and the multi-level structure of AI system design, implementation, and operation processes. The study is based on the concept of Ethical Production Safety Engineering (EPSE), which extends the classic approach to production safety engineering with normative, legal, and ethical dimensions, taking into account the specific nature of autonomous decision-making systems based on artificial intelligence. It is a proposal to move from reactive attribution of blame after dangerous (accidental) events to proactive design of responsibility as an integral part of occupational safety in modern enterprises. The paper presents the theoretical and conceptual foundations of ethical production safety engineering and assumes that this safety is systemic and emergent in nature. The use of AI means that responsibility is treated as a systemic, distributed construct oriented towards the life cycle of the system. Key to this concept is the Multi-Layer Responsibility Allocation Model (MLRAM), which reflects the complex, distributed structure of agency and decision-making in AI-driven production systems. In the presented approach, responsibility is relational, multilevel, and extended over time, covering the entire life cycle of an autonomous system. It is assessed on the basis of sets of layered indicators relating to specific decisions and practices characteristic of each level of the system. The developed solution represents a new approach to assessing the safety status of complex and increasingly autonomous production systems using AI tools.