I wonder how to deal with data that is accidentally identifiable. For example, imagine that you are running an anonymous poll or survey. In the general case that would not identify an individual person, but in some circumstances a particular collected answer will be unique and could theoretically be connected to an individual.

In such cases it's not really possible to give individuals control over their data, because except for the special case the whole point is that it's not connected to an individual...