Except that it has been demonstrated that no amount of 'anonymization' or 'pseudonimization' is effective at actually disconnecting the data from the person. Wi

Except that it has been demonstrated that no amount of 'anonymization' or 'pseudonimization' is effective at actually disconnecting the data from the person. With enough data, you can figure out who's who even with this 'randomID'. Simple example: A random user with ID 'aaa' will provide location data throughout the day. This provides you with the means to figure out where they work, how late they get up, what their average walking speed is, etc. One day the user get's a new randomID 'bbb'. At first glance there is no connection between 'aaa' and 'bbb'. But from a specific day 'aaa' stops transmitting data, and 'bbb' starts transmitting data. 'bbb' will submit the same data and thus you can figure out for example their walking speeds and place of work. This means 'aaa''s data will match 'bbb''s. Looking at the datetime when the switchover happened you could deduce that 'aaa' and 'bbb' are probably the same. With more data points you can increase your accuracy. IE: with sufficiently large datasets (which they have), no amount of anonymization works (which they know).