Recital 33

The development of certain AI systems and AI models may involve the collection of large amounts of data, which can include personal data and special categories thereof. Special categories of personal data may incidentally and residually exist in the training, testing or validation data sets or be retained in the AI system or the AI model, although the special categories of personal data are not necessary for the purpose of the processing and while the controller did not intend to process such personal data and has taken the appropriate technical and organisational measures to avoid such processing, such as the filtering out of special categories of personal data at the collection stage.

In order not to disproportionately hinder the development and operation of AI and taking into account the capabilities of the controller to identify and erase special categories of personal data, including technical capabilities, derogating from the prohibition on processing special categories of personal data under Article 9(2) of Regulation (EU) 2016/679 should be limited to incidental and residual processing of special categories of data in the context of the development and technical operation of AI systems or AI models. The derogation should not be understood as covering the processing of special categories of personal data directly acquired by or provided to the AI system or AI model in the course of the deployment. The derogation should only apply where the controller has implemented appropriate technical and organisational measures in an effective manner to avoid the processing of those data, takes the appropriate measures throughout the entire lifecycle, that is to say during the development and operation, of an AI system or AI model and, once it identifies such data, effectively erase them. If erasure would prove technically impossible or require manifestly disproportionate effort, notably where the erasure of special categories of data memorised in the AI system or AI model would require re-engineering the AI system or AI model, or would be technically impossible, the controller should effectively protect such data from being further processed or processed for other purposes, in particular being used to infer outputs, being disclosed or otherwise made available to third parties. In line with the accountability principle, the controller should document its assessment and have processes in place to monitor and demonstrate the effectiveness of these measures. This derogation should not apply where the processing of special categories of personal data is necessary for the purpose of the processing for example if special categories of personal data relating to a natural person’s voice, image or other personal characteristics are used in a way that goes beyond incidental or residual existence. In this case, the controller should rely on the derogations pursuant to Article 9(2)(a) – (j) of Regulation (EU) 2016/679 or on other Union law, such as Regulation (EU) 2024/1689 regarding the processing of special categories of personal data for the purpose of ensuring bias detection and correction. The notion of AI system and AI model should be understood in the same manner as in Regulation (EU) 2024/1689.