Artificial intelligence is creating new opportunities to shift workplace safety and health from reacting to incidents toward anticipating and preventing harm. This presentation explores how AI can support this transition across the safety lifecycle. Computer vision can help detect unsafe conditions and recurring patterns, while predictive models may provide earlier warning of developing risks. Moving further upstream, AI can also support Design for Safety by combining project information with structured safety knowledge to identify risks and consider appropriate controls before hazards reach the worksite. However, effective Safety AI requires more than technical accuracy. It must provide traceable evidence, support professional judgement and lead to meaningful safety action. The presentation highlights emerging research and practical examples, while emphasising that AI should strengthen, rather than replace, human responsibility and safety culture.