Source: EDPB Guidelines 02/2026 on Anonymisation (draft)
Draft EDPB guidelines (July 2026) replacing the 2014 test with three criteria (no isolation, no linkage, no inference) and a relative, recipient-aware notion of anonymity that explicitly covers AI model outputs.
Contents
- What: draft guidelines (version 1) adopted for public consultation
- Publisher: European Data Protection Board
- Date: adopted 2026-07-07; consultation closed 2026-10-30; final text pending as of 2026-11-14
- Original: EDPB PDF. Content summarised from the Stephenson Harwood analysis; the PDF itself was not read.
Summary
Updates the EU anonymisation standard. Anonymity is relative to who holds the data and what means they reasonably have, following the CJEU’s EDPS v SRB judgment (C-413/23 P). Anonymisation is itself a processing activity that must be documented and tested.
Key claims
- Replaces Article 29 Working Party Opinion 05/2014.
- Three criteria; data meeting all three is presumed anonymous: no record isolation, no linkage to other datasets, no meaningful inference about an individual. Failing one criterion does not automatically make data personal; it triggers further analysis (per Sidley).
- Two approaches: contextual (per-recipient capabilities) and simplified (more cautious, ignores recipient differences); they can be combined.
- The same dataset can be personal for one entity and anonymous for another.
- Meaningful inferences from prompting AI models, and synthetic datasets, can breach the no-inference criterion, so model outputs are not anonymous by default; agentic AI may cut the cost of re-identification.
- The same EDPB batch included Guidelines 03/2026 on web scraping for generative AI.