Health data is one of AI's hardest trust problems: patients want its benefits, but not at the cost of privacy or consent. This talk introduces the Australian Institute for Machine Learning and the three pillars underpinning AI progress: data, energy and algorithms. It then explores two frontiers from AIML's own research, machine unlearning and multi-modal medical AI, before pivoting to real-world impact: health data trials running across South Australia right now, and the Australia AI Hub. The result? A grounded case for what responsible AI in health actually looks like in practice.