Our work spans building new AI models for health-related challenges and studying how deployed systems behave and fail in safety-critical contexts. We take an integrated approach in which AI safety, model development, and real-world deployment in health and public health are studied together, creating a feedback loop between methods, models, and impact.