The authors of this research should really explain better to journalists what the difference between anonymized and de-identified data is. Their paper does a go

The authors of this research should really explain better to journalists what the difference between anonymized and de-identified data is. Their paper does a good job at this and does not claim that anonymization is impossible, most of the newspaper articles that were published about it seem to miss this point entirely though. To clarify, when de-identifying a dataset you simply remove direct identifiers (like a name) from it. This protects the data from direct re-identification, i.e. from someone learning the identity of a person in the data by looking at an individual row. Anonymization is supposed to protect individuals from re-identification also when using external context information and can usually only be achieved by further transforming the data, for example by grouping it (as techniques like k-anonymity do), adding noise (as randomized response techniques do) or by synthesizing new data. High-dimensional, de-identified data will always be easy to re-identify given enough context, I've done this myself in 2016 with a clickstream dataset (the authors reference our work in their paper).