Truly Anonymous? The EDPB’s Guidance on Data Anonymisation

Many organisations rely on anonymisation to enable the broader use, sharing and analysis of data for purposes such as research, product development and AI training, as genuinely anonymous information falls outside the scope of the EU General Data Protection Regulation (GDPR). Deciding whether data is genuinely anonymous, however, is not as straightforward as it seems, and an organisation that gets it wrong may be processing personal data without meeting its GDPR obligations, opening itself to the liabilities that follow non-compliance. In this blog post, we discuss the European Data Protection Board’s (EDPB) draft Guidelines 02/2026 on Anonymisation and examine how organisations can distinguish data that is genuinely anonymous from data that has merely had its obvious identifiers removed. 

 

The EDPB’s Draft Guidelines 

On 7 July 2026, The EDPB adopted draft Guidelines 02/2026 on Anonymisation.. This Guideline updates the Article 29 Working Party’s Opinion 05/2014 on anonymisation techniques (WP216), which has been the main reference point for more than a decade. The EDPB explains that an update was needed because of developments in the case law of the Court of Justice of the European Union (CJEU), the creation of EU-wide data spaces, and advances in AI and technology generally. 

Th EDPB had published the drafts for public consultation until 30 October 2026. Following the consultation, the EDPB is expected to adopt a final version, which may differ from the draft. EDPB guidelines are not legally binding in themselves. They do, however, set out how Europe’s data protection authorities collectively read the law, which makes them an important indication of how regulators are likely to approach the question.  

 

Anonymous Data and the Limits of “Removing Identifiers” 

The GDPR applies to personal data, meaning information relating to an identified or identifiable natural person (Article 4(1)). Recital 26 explains that, to decide whether a person is identifiable, account should be taken of all the means “reasonably likely to be used” to identify them, either by the controller or by another person. Information that does not relate to an identified or identifiable person is anonymous, and the GDPR does not apply to it. 

Data does not have to be impossible to re-identify to count as anonymous. According to the CJEU, the question is whether identification is a realistic risk, not merely a theoretical possibility. If identifying someone would be illegal, or would take so much time, cost and effort that it would not realistically happen, that route to identification is not taken into account (Breyer, C-582/14; EDPS v SRB, C-413/23 P). 

What this test asks is whether a person can be identified, not whether their name has been deleted. Removing names, email addresses, or ID numbers does not by itself make data anonymous. The draft Guidelines illustrate this with a simple example. On a busy street, many people wear black suits and many people walk dogs, so neither description identifies anyone. If only one person is wearing a black suit while walking a dog, however, that combination singles them out. Datasets work in the same way: details that are harmless on their own may, together, point to one person. 

 

Assessing Anonymity 

The sequence for assessing anonymity below draws on the framework in the draft Guidelines. 

To assess organisations should consider: 

  1. Does the Information Relate to a Person?

Information relates to a person where, because of its content, purpose, or effect, it is linked to them. This can include information that is mainly about an object, such as a car or a house, where it is used to evaluate or affect the person connected to it. The EDPB notes that even aggregated statistics may relate to individuals where techniques exist to extract information about them. 

  1. For Whom Should the Data Be Anonymous?

Because anonymity can depend on perspective, the EDPB recommends identifying every party for whom the data is meant to be anonymous, as well as anyone who could realistically gain access to it. Depending on the circumstances, this may include the organisation itself, recipients, rogue employees, cybercriminals, or investigative journalists. 

The draft Guidelines make two further points. First, a service provider that processes data on behalf of another organisation should assess the data from that organisation’s perspective. It therefore cannot treat data as anonymous simply because it holds no identifying information itself. Second, a recipient’s means include means available through others, for example where it could easily hire a third party to carry out re-identification. 

  1. Does the Data Pass the Three Criteria?

The EDPB updates the three risks identified in WP216 (singling out, linkability and inference) into three criteria: 

  • No Record Isolation: no record in the dataset contains a unique combination of details relating to a single person; 
  • No Linkage: the data cannot be linked with other information about the same person; and 
  • No Inference: no specific and meaningful conclusion about an individual can be drawn from the data. 

If all three criteria are met, the EDPB considers that the data can be regarded as anonymous. If one is not met, the data is not automatically personal data, but further analysis is needed. 

The draft Guidelines propose two ways to apply the criteria. The contextual approach assesses the actual capabilities of each relevant party and reflects the full legal test. The simplified approach ignores differences between parties and assumes that if a re-identification method exists, someone will use it. The simplified approach is more cautious and easier to apply, but it may lead an organisation to treat data as personal even where it would be anonymous for some recipients. The EDPB stresses that the simplified approach is not an alternative legal standard, and that the two approaches can be combined. 

  1. Has the Anonymisation Itself Complied with the GDPR?

According to the EDPB, anonymising personal data is itself a processing activity. It therefore requires a legal basis under Article 6 GDPR and, for special category data such as health data, an exception under Article 9(2). In the EDPB’s view, this legal basis should be presumed to be the same as the basis for the original processing where the anonymisation forms part of the same activity and serves the same purposes. Organisations should also be transparent about the anonymisation, avoid describing data as “anonymous” if people remain identifiable, and document both the process and the testing. 

The EDPB also notes that a contractual ban on re-identification is not the same as a prohibition by law. Contracts can complement technical measures, but they cannot replace them. 

  1. Does the Conclusion Still Hold?

Because anonymity can change over time, the EDPB describes periodic reassessment as good practice. It also clarifies that organisations that assessed data as anonymous under WP216 before the draft Guidelines were published are not expected to carry out a new assessment. 

 

What About AI? 

AI does not change the legal test, but it is one of the main reasons the answer may change over time. The EDPB observes that re-identification techniques are becoming more accessible and that AI, especially agentic AI, is likely to reduce the time and cost of using them. The draft Guidelines also note that AI models and synthetic data can sometimes be queried or prompted to reveal information about individuals in the underlying data. 

In our view, this makes the practical assessment of identifiability more important for organisations that use anonymised data for AI development, analytics or data sharing. 

 

Key Takeaways 

The draft Guidelines do not make anonymisation unachievable, but in our view they shift the question organisations should ask. Rather than Have we removed the identifiers?, the question becomes: Can an individual still reasonably be identified from this information, in these circumstances, using the means reasonably likely to be used? 

Organisations that rely on anonymisation will need to be able to show how they reached that conclusion. One practical approach could be to: 

  • map who will receive or could access the data; 
  • test the data against the three criteria; 
  • record the reasoning; and 
  • revisit the assessment when circumstances change, for example when the data is shared more widely or new datasets become available.