Unique threats to patient safety and well-being from artificial intelligence (AI) in health care applications, such as bias, are now well-recognized. At the same time, policy makers and regulators are grappling with the challenges of evaluating and regulating AI-based technologies in order to minimize these threats. Some potential harms arise from characteristics inherent in the technology, such as the “black box” nature of some machine learning, or the limitations of using AI to infer causation. Others arise from messy, incomplete, or unrepresentative datasets. Some of these challenges can be addressed technologically, while for others these ‘fixes’ may exacerbate the biases they intend to prevent. Others, still, will be reliant on systemic changes. This panel will discuss innovative regulatory, policy, and other approaches to minimizing harms to patients from health care AI.