The Belgian authority orders a chatbot application to comply with the GDPR regarding sensitive data

This decision by the Belgian Data Protection Authority (DPA) analyzes the applicable legal basis for training artificial intelligence models, distinguishing the processing of standard personal data from that of sensitive data.

Facts and context

The Data Protection Authority (DPA) today published a decision ordering an online chatbot application to comply with the GDPR, due to breaches related to the legal basis for processing sensitive data and lack of transparency.

The case originated from a complaint concerning the alleged use without legal basis of conversational data for training the application's artificial intelligence models.

Reasons for the decision

  • Obligation to have a legal basis for processing special categories of data (Article 9 of the GDPR): The authority distinguished "normal" conversational data from sensitive data. While it accepted that the legitimate interest of the controller could justify processing non-sensitive data to optimize its model, it found this legal basis insufficient for sensitive data. For the latter, which includes information about health or sexual orientation, the GDPR imposes a general prohibition, lifted only in limited cases such as the explicit consent of the data subject, which was lacking in this case.
  • Transparency obligation (Article 12 of the GDPR and Article 13 of the GDPR): The authority found that the company did not provide sufficiently clear and complete information about the use of conversational data, whether normal or sensitive, for training its artificial intelligence. Although changes were made to the privacy policy during the procedure for non-sensitive data, the information regarding the processing of special data remained incomplete.

Decision of the authority

Consequently, the authority issued a reprimand to the company.

Furthermore, the authority ordered the company to bring the processing of special conversational data and the information provided to users about it into compliance within 4 months. This injunction is accompanied by a penalty of €1,000 per infringement per day of delay, totaling €2,000 per day.

Lessons learned

This decision clarifies that:

  • legitimate interest can constitute a valid legal basis for training an artificial intelligence model with non-sensitive personal data, provided the processing is necessary, expected by users, and accompanied by appropriate safeguards.
  • processing sensitive data for training artificial intelligence models imperatively requires one of the exceptions provided in Article 9, paragraph 2 of the GDPR, such as explicit consent, legitimate interest being generally insufficient.
  • transparency requires clearly distinguishing in the information provided to data subjects the purposes and legal bases applicable to the processing of standard data from those applicable to sensitive data.

Informations complémentaires

L'analyse complète est réservée aux membres

Montant de la sanction, thèmes, secteurs, entités et données concernées : l'essai gratuit de 14 jours ouvre la fiche entière et la veille personnalisée.

Essayer gratuitement 14 jours · accès complet · sans carte bancaire