Worried about Federated Learning of Cohorts (FLoC) particularly wrt topics that might be actionable. Will there be a cohort that incorporates people who visit abortion websites, for example? How actionable might that data be in an adverse regime?