What counts as scientific research?

The EU’s General Data Protection Regulation 2016/679 (“GDPR”) applies also where personal data is processed for scientific research purposes. However, the GDPR does not formally define “scientific research”, and so far, there has not been a clear consensus on the definition of the term under the GDPR.1 The distinction matters, as processing personal data for scientific research purposes benefits from specific rules and flexibilities under the GDPR, including, for example, presumed compatibility of further processing and the possibility of longer retention under Article 5, subject to Article 89 safeguards. Article 17 also provides a conditional exception to erasure where erasure would make research impossible or seriously impair its objectives.

According to the recitals of the GDPR, scientific research may also be privately funded.2 Recital 159 supports a broad interpretation of scientific research, including, inter alia, privately funded research, technological development and applied research. Consistently with that approach, the European Data Protection Board (“EDPB”) recognises that scientific research may also aim to further commercial interests. Therefore, commercial use of the research results does not, in itself, preclude the rules regarding scientific research under the GDPR to apply.3 However, if the research does not itself qualify as scientific research in the meaning of the GDPR, the associated processing of personal data cannot be considered to be conducted for scientific research purposes.4

The EDPB has published Guidelines 1/2026 to clarify the concept of processing of personal data for scientific research purposes. EDPB underlines in its Guidelines 1/2026 that research needs to be genuinely scientific in order to be covered by the GDPR’s rules for processing data for scientific research purposes.5 The EDPB identifies six key-indicative factors to support in determining whether processing of personal data is carried out for scientific research purposes:

  • Methodical and systematic approach: following an appropriate methodology and research plan, testing a hypothesis or, for exploratory work, pursuing a stated objective to a hypothesis.
  • Adherence to ethical standards: conducting the research activities in accordance with the relevant field’s ethical standards.
  • Verifiability and transparency: producing verifiable results open to scrutiny, normally through peer review, and sharing results, while respecting legitimate intellectual-property and trade-secret limitations.
  • Autonomy and independence: enabling appropriately qualified researchers, or those under their supervision, to work free from undue pressure and make scientific choices independently.
  • Research objectives: aiming to contribute to society’s general knowledge and wellbeing, even where commercial interests are also pursued.
  • Scientific contribution or novelty: having the potential to contribute to existing scientific knowledge or apply it in novel ways.6

Meeting all six factors supports a presumption that the activity is scientific research, in the meaning of the GDPR. However, missing a factor does not automatically exclude the activity from the definition, but it the controller must justify and demonstrate why it nevertheless qualifies as scientific research. Therefore, the assessment remains contextual, rather than a fixed set of criteria.7

Would training an AI model constitute scientific research?

Potentially, but not automatically. The question is whether the particular training activity forms part of genuinely scientific work, not merely whether it uses advanced technology. A documented project testing a new training method, with ethical safeguards, independent scientific judgement and verifiable results, would have a stronger claim than routine product optimisation undertaken solely to improve commercial performance.8

The EDPB’s example in the Guidelines 1/2026 where a generative-AI startup and a university research model bias using established methods, ethical review and peer-reviewed publication is instructive. The project qualifies as scientific research even though the startup may exploit the results commercially. This supports the conclusion that commercial AI development is not excluded, but not every commercial AI-training activity is scientific research. Trade secrets alsodo not automatically disqualify a project as scientific research, although legitimate confidentiality must be reconciled with verifiability and sharing results.9

Failure to qualify does not mean AI development is necessarily unlawful. If the activities do not qualify as scientific research under the GDPR, the specific provisions governing scientific research do not apply, but the processing of personal data may be justified based on other provisions. In its recommendations on AI-system development, the data protection authority for France (Commission Nationale de l’Informatique et des Libertés, “CNIL”) identifies legitimate interests under Article 6(1)(f) as a possible legal basis for developers. This requires a lawful, specific interest, necessary processing and a balancing assessment protecting individuals’ rights and reasonable expectations. Other GDPR requirements also remain fully applicable, including, for example, a separate Article 9 condition where special-category data are processed.10

More guidance, but not the final word

The Guidelines 1/2026 considered here are EDPB’s non-final public-consultation version for which the public consultation ended on 25 June 2026. They provide welcome direction, but a more common understanding and consensus of the definition of scientific research under the GDPR will depend on the finalised guidelines and future case law.

1Finnish case law illustrates the difficulty of drawing the boundary with the definition of scientific research. In case KHO:2013:181, the Supreme Administrative Court recognised that pharmaceutical-company funding did not, in itself, prevent a study from being scientific research. However, it emphasised autonomy and publicity and upheld refusal of access to register data because the study’s primary scientific aims had not been sufficiently established, taking account of sponsor influence and limitations in the study design. The decision concerned pre-GDPR Finnish rules on access to confidential official records, not the GDPR definition of scientific research.

2Recital 159 of the GDPR.

3EDPB, Guidelines 1/2026 on processing of personal data for scientific research purposes, adopted 15 April 2026, public-consultation version, paragraph 11 (v) and Example 2.

4EDPB, Guidelines 1/2026 on processing of personal data for scientific research purposes, adopted 15 April 2026, public-consultation version, Example 3.
5EDPB, Guidelines 1/2026 on processing of personal data for scientific research purposes, adopted 15 April 2026, public-consultation version, paragraph 9.

6EDPB, Guidelines 1/2026 on processing of personal data for scientific research purposes, adopted 15 April 2026, public-consultation version, paragraph 11 (i.-vi.) and Example 2.

7EDPB, Guidelines 1/2026 on processing of personal data for scientific research purposes, adopted 15 April 2026, public-consultation version, paragraph 12.

8EDPB, Guidelines 1/2026 on processing of personal data for scientific research purposes, adopted 15 April 2026, public-consultation version, paragraph 11 and Example 2.

9EDPB, Guidelines 1/2026 on processing of personal data for scientific research purposes, adopted 15 April 2026, public-consultation version, paragraph 11 (iii.) and Example 2.

10CNIL, “AI system development: CNIL’s recommendations to comply with the GDPR”, 5 January 2026, Step 3; GDPR, Articles 6(1)(f) and 9. https://www.cnil.fr/en/ai-system-development-cnils-recommendations-to-comply-gdpr.

Our team is happy to advise on any questions you may have about GDPR, AI and healthcare regulation in Finland.

Contact