Evaluating User Safety Aspects of AI-Based Systems in Industrial Occupational Safety: A Critical Review of Research Literature.

AI technologies are becoming increasingly prevalent in industrial workplaces, extending their applications beyond productivity to critical areas such as occupational safety. From our perspective, it is important to consider the safety of these AI systems for users already at the research and development stage, rather than only after deployment. Therefore, in this review, we synthesize publications that propose such AI-based safety systems to assess how potential risks are addressed early in their design and prototype stages. Consequently, we explore current advancements in AI-driven, sensor-based, and human-centered applications designed to enhance occupational safety by monitoring compliance, detecting hazards in real time, or assisting users. These systems leverage wearables and environmental sensing to proactively identify risks, support decision-making, and contribute to creating safer work environments. In this paper, we categorize the technologies according to the sensors used and highlight which features are preventive, reactive, or post-incident. Furthermore, we address potential risks posed by these AI applications, as they may introduce new hazards for workers. Through a critical review of current research and existing regulations, we identify gaps and propose key considerations for the safe and ethical deployment of trustworthy AI systems. Our findings suggest that in AI- and sensor-based research applications for occupational safety, some features and risks are considered notably less than others, from which we deduce that, while AI is being increasingly utilized to improve occupational safety, there is a significant need to address regulatory and ethical challenges for its widespread and safe adoption in industrial domains.