Artificial intelligence (AI) systems are rapidly transforming the workplace, performing tasks once limited to human intelligence such as decision-making, prediction, and pattern recognition. While AI adoption offers opportunities to improve productivity, it can also create new occupational hazards and alter working conditions in ways that may harm worker health, safety, and wellbeing. Despite broader and growing attention to safe and responsible AI, there is limited integration of occupational health and safety (OHS) principles into AI design and adoption decisions. This paper outlines a framework for embedding an OHS perspective throughout the AI system life cycle, from problem definition to system retirement. The framework aims to ensure that safety, fairness, and worker wellbeing are prioritized in AI. We describe key OHS goals for each phase of the AI life cycle and describe practical strategies to support implementation. These strategies include participatory co-design with workers, equitable data collection, model training and validation that identify and minimize safety risks, transparent deployment practices, and continuous monitoring and retraining guided by risk management frameworks. We emphasize collaboration among AI system developers, OHS professionals, and worker and workplace representatives, to anticipate and address emerging risks. Integrating OHS principles into the AI system life cycle not only helps prevent harm but also fosters worker trust, strengthens system reliability, and promotes sustainable technological adoption. Embedding OHS principles into AI development ensures that the technology contributes to, rather than compromises, the protection and wellbeing of workers in a changing world of work.