We use cookies to personalise the website and offer you the greatest added value. They are, among other purposes, used to analyse visitor usage in order to improve the website for you. By using this website, you agree to their use. Further information can be found in our data privacy statement.



The EDPB's new anonymisation guidelines: a practical reset for a twelve-year-old standard

​​​​​​​​​​​published on 17 July 2026 | reading time approx. 5 minutes​​​

​
On 7 July 2026, the European Data Protection Board (“EDPB”) adopted Guidelines 02/2026 on Anonymisation (Version 1.0) (the “Guidelines”), opening the text for public consultation. The document is not yet final: stakeholders can submit comments until 30 October 2026, and the framework described below may still change before formal adoption. That caveat matters, because these guidelines are set to replace the reference point every EU privacy practitioner has relied on since 2014 – the Article 29 Working Party's Opinion 05/2014 on anonymisation techniques (the “2014 Opinion”).

Why now?​​

The 2014 Opinion did solid work for its time, but it was written before the Court of Justice of the European Union’s most influential rulings on identifiability, before the EU's data-space strategy took shape, and well before generative AI made re-identification a matter of prompting rather than programming. The EDPB is explicit about the trigger: case law developments (notably Breyer, EDPS v SRB, Gesamtverband Autoteile-Handel v Scania, and OC v Commission), the emergence of EU-wide data spaces, and advances in AI and re-identification tooling have all exposed gaps in the old three-criteria model. Rather than scraping that model, the EDPB updates and sharpens it.

Importantly, the EDPB clarifies this is not retroactive: Controllers who assessed a dataset as anonymous under the 2014 Opinion are not obliged to redo that assessment – though periodic reassessment is recommended as good practice, since re-identification risk tends to grow over time as techniques and available auxiliary data evolve.

The legal core: two questions, many perspectives​​

The Guidelines restate the GDPR (aka EU Regulation 2016/679) test for anonymity as two sequential questions: does the information relate to a natural person, and, if so, is that person identified or identifiable? A "no" to either question means the data is anonymous – but critically, the answer is not absolute. It must be assessed from the perspective of each relevant entity, since the same dataset can be anonymous for one recipient and personal data for another, depending on what means of identification that entity could reasonably use.

This relativity is the Guidelines' most operationally significant point. "Means reasonably likely to be used" is not a fixed technical threshold; it depends on cost, time, technology, legal barriers, and – crucially – the realistic (not merely theoretical) chain of access an entity has to the data, including through third parties. The EDPB also warns against relying on an entity's presumed "lack of motivation" to identify someone, since motivation is hard to prove and can change.

The EDPB's position: three criteria, two approaches​

The technical heart of the Guidelines is a three-criteria test, refining rather than replacing the model of the 2014 Opinion:
  • No record isolation – no unique combination of attributes singles out one individual's record;
  • No linkage – no record can be reliably matched to a record about the same person in another dataset;
  • No inference – no specific and meaningful conclusion about an identified or identifiable individual can be drawn from the data.

If all three are met, the data can be treated as anonymous; if any fails, further analysis is needed – in particular, whether isolated or linked records actually allow singling out a real individual, rather than merely looking unique in the abstract.

The EDPB offers two ways to run this test. The contextual approach applies to the full legal standard, weighing each relevant entity's actual capabilities – more precise, but demanding, since it requires mapping who could plausibly access the data and what they could do with it. The simplified approach ignores those differences and assumes that if a re-identification method exists in theory, someone will use it. It is more conservative – it can lead a Controller to treat data as personal even where it would, in fact, be anonymous for some recipients – but it is easier to apply and gives a higher-confidence floor. The two can be combined: start simplified, then refine with context where it helps demonstrate genuine anonymity.

What this means for Controllers​​

For businesses acting as Controllers, the Guidelines translate into a concrete due-diligence checklist to understand how to put the concept of anonymization into practice:
  1. Map the data and its audience. Identify every entity – internal teams, business partners, third party providers, external contractors – who might receive or access the "anonymised" output, since anonymity must hold from their vantage point too, not just the Controller's own;
  2. Test, don't assume. Run the three criteria against the dataset, considering realistic auxiliary information (public registries, leaked datasets, social media, AI-assisted inference) that any relevant entity could plausibly obtain;
  3. Treat aggregation and pseudonymisation with caution. Aggregate statistics can still leak individual-level facts through inference (e.g., outlier salaries, small denominators); pseudonyms and encryption are reversible by design and do not equal anonymisation unless decryption is genuinely out of reach;
  4. Document everything. The anonymisation process, the perspectives considered, and the testing methodology should be recorded and retained – this evidence underpins both the GDPR-compliance of the anonymisation activity itself (which still needs a lawful basis) and any future defence of the "anonymous" label;
  5. Reassess periodically. Re-identification risk is not static; technological progress (especially AI-driven techniques) can turn yesterday's anonymous dataset into today's personal data. A security incident affecting confidentiality assumptions should specifically trigger a reassessment;
  6. Avoid mislabelling. The EDPB is pointed on this: Controllers should not describe data as "anonymous," "de-identified" or "de-personalised" if individuals remain identifiable – a transparency obligation with real enforcement teeth.

The bottom line​

The Guidelines don't lower the bar for anonymisation – if anything, they formalise a more rigorous, perspective-sensitive way of clearing it. For organisations relying on "anonymous" datasets to sidestep GDPR obligations, the message is clear: anonymisation is not a one-off technical fix but an ongoing, documented risk assessment tied to who can access the data and what they could realistically do with it.

Given the guidelines remain in public consultation until 30 October 2026, Controllers have a window to review their own anonymisation practices against this emerging standard – and, if they wish, to contribute to shaping its final form.​

Tech & Data Bites

author

Contact Person Picture

Tommaso Mauri

Attorney at Law (Italy)

Senior Associate

+39 02 6328 841

Send inquiry

Profile

Skip Ribbon Commands
Skip to main content
Deutschland Weltweit Search Menu