I believe lack of privacy preserving models/ systems in machine learning is also a cause for concern. In many applications we collect user data (the biggest companies in the world do that) if we are able to train models in a more secure manner(not just federated learning) people might be more inclined to give more info if they can control and we can guarantee no party apart from the user will get a peek at the raw data.

This is my two cents on the topic.