Anonymisation in 2026: What the European Data Protection Board (EDPB) is Really asking for

Anonymisation in 2026: What the European Data Protection Board (EDPB) is Really asking for

Anonymisation in 2026: What the European Data Protection Board (EDPB) is Really asking for

Understanding the EDPB's Draft Guidelines 02/2026 and what they mean for your organization

On 7 July 2026, the European Data Protection Board (EDPB) adopted draft Guidelines 02/2026 on 'Anonymization.' This is the first time that such guidance has been updated in over ten years by the European Union. Organizations processing data that are considered anonymous will have to now adhere to serious implications as per the renewed guidelines. Whether conducting AI model training, sharing research datasets, conducting analytics, or managing customers' data, the framework is a game-changer when it comes to anonymization and its assessment.

The draft guidelines remain  open for consultation until 30 October 2026 and will be adopted once the guidelines are finalized. However, given the fact that there have been some landmark rulings by EU courts on the framework, it is important to understand what the EDPB is here to bring out.

The Shift: A Contextual and Entity-Specific View of Identifiability 

A central feature of the draft guidelines is their contextual approach to identifiability. With such an approach, the same dataset can be personal data for one company and anonymous for another depending on who holds it and what can actually be done with it.

It is not merely theoretical; the  case related to ‘EDPS v SRB’ (September 2025) showed that pseudonymized data is not always personal data for all recipients. It follows that anonymization cannot be considered a once-and-for-all procedure. A dataset can be completely anonymous for one company but personal data for a third party (if this party has the capacity to restore the identity of the individuals), can still be personal data, and this was a concern.

When data is disclosed or made available to another organization, the role of the recipient needs to be considered. In cases where the recipient is independent and acts on their own purposes, it might be necessary to assess the realistic possibilities of identification of that recipient. On the contrary, when the recipient is a cloud provider, who performs processing solely on behalf of the controller, the position of the data recipient will be assessed as the position of the controller.

Two Ways to Assess Identifiability

In the digital era, there are two ways to assess identity information; the ‘Simplified approach’ and the ‘Contextual approach’. Here both approaches are unique and have their own challenges in so far as its anonymization is concerned. It is also pertinent to mention that the ‘Contextual approach’ is the one that EDPB has endorsed.

Simplified approach: asks solely if re-identification is technically possible by anyone without considering whom the data belongs to and what the actual holder may do. This is a conservative approach; it can identify personal data that actually are anonymous in reality for the actual participants in the process.

Contextual approach: the approach that  EDPB endorses and is consistent with EDPS v SRB. First, identify who should have the anonymized data, and then check if these actual recipients have the means "reasonably likely to be used" to re-identify the data. That is when the relative identifiability concept becomes operational, same dataset, different conclusions depending on the recipient.

Anonymization Is Processing and must Comply with the GDPR 

An important legal aspect that many organizations ignore: the act of anonymization itself counts as processing under the GDPR. Thus, organizations cannot conduct anonymization activities without having a legitimate ground (Article 6) and, if special categories of data are processed, an appropriate derogation (Article 9(2)).

This legal need changes the way organizations justify their anonymization projects. The fact that the data obtained after processing will be anonymous cannot be used as grounds for ignoring GDPR legal obligations when conducting anonymization activities. In cases when data is anonymized for research, AI training, or analytics, the legal basis for the anonymization activity should be identified prior to any processing.

Organizations that implement large-scale anonymization projects, especially in areas of AI, healthcare, and analytics, have an obligation to audit their legal bases and processing justifications. In case anonymization activities are conducted without legal justifications, such an issue should be addressed first.

Documentation and Periodic Reassessment: Accountability in Practice 

The draft guidelines place significant emphasis on retaining documentation that shows how the organization assessed and implemented anonymisation. It includes the following: 

  • Technical specifications of how anonymisation was performed and which techniques were applied

  • Risk assessments demonstrating irreversibility and unlinkability

  • Testing results of supposedly anonymous datasets (e.g., re-identification testing and validation)

  • Protocols for periodic reassessment as technology and data availability landscape evolve

Documentation must be retained even after the anonymisation is finished to support GDPR compliance and the efficiency of anonymisation procedures.

Periodic reassessment deserves special attention as well. According to EDPB, the static assessment of the dataset anonymity should be avoided, especially where the data remains in circulation and additional third-party data can be obtained. The companies should determine the frequency of reassessments according to the sensitivity of the data and its usage pattern and technology advancement rate.

Organisations that keep anonymised datasets in use, need more than just a static assessment to meet regulatory requirements. Periodic reviews of the efficiency of anonymisation procedures need to be done according to the technology advancement and third-party data availability changes. It is a resource-demanding task, however, it cannot be avoided.

Three Criteria Framework: No Record Isolation, No Linkage, No Inference

The proposed guidelines propose a more polished three criteria approach, pointing out that anonymization works only if all three of the following criteria are met:

  • No Record Isolation :there is no combination of attributes within the data which will isolate the record of one individual. Larger records are less likely to meet this condition.

  • No Linkage: an individual’s record cannot be reliably linked to another record concerning the same individual in a different database (for instance, by using matching identifiers or attributes).

  • No Inference: no reliable inference can be made about an identifiable individual through the use of the data. The EDPB is very precise in this, it includes attribute inference, deriving record data from aggregate data, membership inference, and inferential linkage (including inference by AI model prompting).

Any approach to anonymization, which meets only one or two of the criteria described above, doesn’t fulfill the requirements of the new framework. It is important to test your anonymization techniques (pseudonymization, aggregation, generalization, differential privacy) against all three criteria. Security through obscurity and limited access to some datasets will most likely fail the relative identifiability test.

Implications for Organisational Practice

1. Conduct reassessment of anonymised datasets in use: Although organizations do not have to perform immediate reassessments for all datasets anonymized in accordance with the 2014 Opinion, the best practice involves the periodic reassessment of datasets currently in use or shared by an organization. This step is especially necessary when data is highly sensitive or patterns of disclosures have changed.

2. Develop solid documentation process: Organizations should create comprehensive documentation on the anonymization process (technical methods of anonymization, results of testing, risk assessments, reassessment schedule etc.)

3. Identify risks of re-identification from the point of view of the recipient: Where data sharing takes place, it is necessary to analyze re-identification risks from the perspective of the recipient organization, identifying data that is accessible by them and the ways of its re-identification.

4. Assess legal bases for anonymization: Organizations should make sure that the legal basis for anonymization (and not for subsequent use of anonymized data) has been documented. If an organization conducts anonymization for the purposes of research, artificial intelligence training or analysis, Article 6 compliance needs to be proved from the beginning of the anonymization process.

5. Oversight of guidance completion: The consultation ends on 30 October 2026. Though not yet finalised, the EDPB guidance is unlikely to deviate significantly. Organisations who prepare in advance, by adjusting practices before final publication, will be better placed to comply in the future.

Conclusion: What This Means for Compliance

In its draft guidance, the EDPB recognises the evolution of regulation towards anonymisation being an exact science, rather than an easy compliance task. Anonymisation now requires thorough documentation, reassessment, and a discussion of possible risks of re-identification.

From the perspective of sharing information, training AI systems, or researching with large data sets, the implications of regulation are clear: current anonymisation processes will be increasingly scrutinised. Documentation, reassessment as part of regular processes, and anonymisation evaluation from a data recipient's perspective should be seen as key compliance aspects.

Though the guidance is still subject to consultations, the enforcement reality makes sure that key elements of the framework will stay intact. Organisations who prepare for the above obligations proactively will find compliance easier.

Key Sources

EDPB Guidelines 02/2026 on Anonymisation (Draft, July 2026)   https://www.edpb.europa.eu/system/files/2026-07/edpb_guidelines_202602_anonymisation_v1_en_0.pdf 

EDPB Public Consultation on Guidelines 02/2026

https://www.edpb.europa.eu/public-consultations/guidelines-022026-on-anonymisation_en 

Stephenson Harwood: The EDPB's Guidelines on Anonymisation https://www.stephensonharwood.com/insights/the-edpbs-guidelines-on-anonymisation-one-legal-standard-two-approaches-three-criteria/ 

Sidley Data Matters: EDPB Publishes Draft Guidelines on Anonymisation https://datamatters.sidley.com/2026/07/27/edpb-publishes-draft-guidelines-on-anonymisation/ 

Inside Privacy: EDPB Publishes Draft Guidelines on Anonymisation https://www.insideprivacy.com/eu-data-protection/edpb-publishes-draft-guidelines-on-anonymisation/ 

IAPP: The EDPB's Draft Anonymization Guidelines https://iapp.org/news/a/the-edpb-s-draft-anonymization-guidelines-what-they-mean-for-your-data-strategy  

Freshfields: Anonymous or Not? The EDPB's New Draft Guidelines https://www.freshfields.com/en/our-thinking/blogs/technology-quotient/anonymous-or-not-the-edpbs-new-draft-guidelines-on-anonymisation-102nbv5 

Privacy Rules: EDPB Anonymisation Guidelines 2026 —https://privacy-rules.com/edpb-anonymisation-guidelines-2026-what-they-change-and-how-to-comply/ 

EDPS v SRB (Court of Justice of the European Union, September 2025)