Skip to content
Atesto
Create account
All articles

Data protection

EDPB adopts final legitimate interest guidelines for AI under GDPR

The European Data Protection Board adopted version 2.0 of its legitimate interest guidelines. Companies using AI must document a strict three-part test.

On 10 November 2026, the European Data Protection Board (opens in a new tab) adopted version 2.0 of Guidelines 1/2024. This document provides final supervisory guidance on processing personal data under Article 6(1)(f) of the General Data Protection Regulation. The adoption follows a public consultation period. The guidelines clarify how the lawful basis test applies to artificial intelligence operations. They confirm that companies do not have blanket permission to scrape personal data or train models simply by claiming a legitimate interest. The official publication includes the adopted document, tracked changes, a consultation report, and a flowchart for the assessment.

The guidelines represent authoritative supervisory guidance on how to interpret the existing regulation. They are not a new European Union regulation, a legislative amendment, or a specific permission for any particular artificial intelligence training dataset. The binding text of the General Data Protection Regulation and the judgments of the courts remain the controlling legal authority. The updated document details exactly what supervisory authorities expect to see when an organization relies on legitimate interest for model training or evaluation.

Who it applies to

The guidelines apply to any data controller that relies on Article 6(1)(f) of the General Data Protection Regulation as its legal basis for processing personal data. This includes artificial intelligence developers building models and organizations processing personal data for these operations. When model training or evaluation includes personal data, the controller must identify an applicable legal basis and comply with the rest of the regulation.

The scope covers various processing purposes, which the guidance treats as distinct operations. Training, fine tuning, model evaluation, fraud detection, security monitoring, and marketing are not interchangeable. A controller must map each purpose separately. The guidelines also clarify that legitimate interest is not a standing exemption for processing publicly accessible photographs, online profiles, or general web content. Public availability does not remove the protections afforded to personal data.

Organizations must not confuse this final adoption with other ongoing regulatory processes. Version 2.0 of Guidelines 1/2024 on Article 6(1)(f) is final. The European Data Protection Board has a separate consultation process for Guidelines 03/2026, which specifically address web scraping for generative artificial intelligence. Those web scraping guidelines were published as a draft and remain separate from this final adoption.

What it requires

Organizations relying on legitimate interest must perform and document a strict Legitimate Interest Assessment. The European Data Protection Board mandates that three conditions must be satisfied cumulatively. Satisfying one part of the test does not cure a failure on another part.

First, the controller must identify a legitimate interest. This interest must be lawful, sufficiently precise, real, and present. The interest can be pursued by the controller or by a third party. The guidelines specify that vague descriptions, such as stating a desire to build better artificial intelligence, are not a substitute for a concrete assessment. The organization must record the precise interest and identify the specific holder of that interest.

Second, the controller must pass a necessity test. The organization must demonstrate why the processing of personal data is necessary for that specific interest. This step requires the controller to evaluate alternatives. The organization must prove whether a less intrusive viable means can achieve the same result. The controller must record the alternatives considered and the reasons for each necessity decision.

Third, the controller must conduct a balancing test. This requires evaluating the interests and fundamental rights of the affected people. Organizations must document the reasonable expectations of the individuals, the sensitivity of the data, the context of the processing, the scale, and the consequences. The controller must document exactly why its own interest is not overridden by the rights of the data subjects.

The assessment must also evaluate mitigation measures based on their actual effectiveness. Safeguards cannot be treated as a formulaic justification to force a passing result on the balancing test. The organization must document concrete and effective safeguards.

Furthermore, the guidelines remind controllers that legitimate interest does not waive other obligations. Processing special category data may require meeting the additional conditions of Article 9. Organizations must also fulfill their information duties under Articles 13 and 14. They must adhere to data minimization, purpose limitation, and retention requirements. They must handle data subject rights and conduct Data Protection Impact Assessments under Article 35 where relevant. The guidance also notes that text and data mining copyright permissions and duties under Article 53 of the Artificial Intelligence Act are separate legal questions.

The dates

The European Data Protection Board adopted version 2.0 of Guidelines 1/2024 on 10 November 2026. Because this document provides interpretative guidance on the existing General Data Protection Regulation, the underlying legal obligations are already in force and applicable. The guidance reflects how supervisory authorities interpret the lawful basis test today.

What to check now

  • Map each processing purpose separately, ensuring that training, fine tuning, and evaluation are not treated as interchangeable operations.
  • Identify the controller, the specific datasets, the categories of individuals, the data types, and any international transfers involved in the operation.
  • Decide and document whether personal data is actually processed in each specific artificial intelligence workflow.
  • Review existing assessments to ensure the stated interest is precise, real, and present, rather than a vague description.
  • Document the alternatives considered for each project and record the specific reasons why less intrusive viable means were rejected.
  • Document the interests and reasonable expectations of the affected people, along with the scale and consequences of the processing.
  • Put workflows for handling data subject rights, including objection handling under Article 21, into the operational process.
  • Revisit the assessment after any significant changes to data sources, processing purposes, or models.

A written Legitimate Interest Assessment serves as evidence of an assessment. It is not a regulatory safe harbor or a guarantee that Article 6(1)(f) applies to the operation. Organizations must secure and document their processing specific balancing exercise to meet the expectations of supervisory authorities.

Sources

  1. European Data Protection Board (opens in a new tab) · euaiactus.com

This article is general information, not legal advice. For your company's situation, talk to a qualified lawyer.

This article was written with AI and checked, sentence by sentence, against the sources linked above.