„We removed the identifiers” can fail you

← All articles
Data protection · EN

„We removed the identifiers” can fail you

My takeways on the EDPB draft Guidelines 02/2026 on Anonymisation (open for consultation until 30 October):

1. The same dataset can be personal data for one entity and anonymous for another. This tracks the CJEU’s reasoning in EDPS v SRB (C-413/23 P): identifiability turns on the means reasonably likely to be used by each relevant entity.

2. A clearer technical test. The EDPB reframes the analysis around three cumulative criteria, no record isolation (singling out), no linkage, and no inference.

3. Membership inference attacks, de-aggregation of statistics, querying a model to elicit new personal data, these are explicitly named as anonymisation risks. If you train or deploy AI, that’s the passage to read twice.

4. Anonymisation is itself processing. It needs an Art. 6 basis and transparency.

The guidelines update the 2014 WP29 Opinion and they read like a document written for the AI era, because they are.

This post has a document attached on LinkedIn. View on LinkedIn →

Discuss on LinkedIn