I follow the research on adversarial machine learning reasonably closely, and have published on things like machine learning supply chains, auditing ML models without access to the training data (using zero knowledge), and how well models can reason about information sharing.