GDPR and artificial intelligence meet on a precise object: the processing of personal data. Regulation (EU) 2016/679 applies, under Article 2(1), “to the processing of personal data wholly or partly by automated means”. Article 4(1) defines such data as “any information relating to an identified or identifiable natural person”; point (2) makes processing any operation performed on it, from collection to destruction.
An AI project that processes such data, in training as in use, encounters six groups of provisions: the principles (Article 5), the legal basis (Article 6), information to individuals (Articles 13 to 15), automated individual decision-making (Article 22), data protection by design (Article 25) and the impact assessment (Article 35).
We read these articles in the consolidated version from the Publications Office’s Cellar repository, consulted on 12 November 2026: none of them uses the expression “artificial intelligence”. This reading does not assess any project: that role belongs to the organisation, its data protection officer and its lawyers.
The principles of Article 5 in the face of a model and its answers
Purpose limitation and data minimisation (paragraph 1, points (b) and (c)) govern the choice of data entrusted to a model: our article on data governance for AI reads them together with Article 6(4).
Point (d) requires data to be “accurate and, where necessary, kept up to date”. Point (e) limits their storage, in a form which permits identification of individuals, to “no longer than is necessary for the purposes”. Under paragraph 2, the controller “shall be able to demonstrate” compliance with these principles.
The article leaves it to be determined what these words cover. Is a generated answer that attributes an inaccurate fact to a person inaccurate data? How long should users’ requests be kept? The answer is written with the data protection officer, and kept.
A legal basis to be found in Article 6
Article 6(1) opens as follows: “Processing shall be lawful only if and to the extent that at least one of the following applies”. There follow consent (point (a)), contract, legal obligation, vital interests, public interest task (points (b) to (e)), and the legitimate interests of the controller or a third party (point (f)). The latter applies “except where such interests are overridden by the interests or fundamental rights and freedoms of the data subject”.
The article does not designate any of these conditions for training or using a model, and we do not designate any. Two publications by authorities, consulted on the same day, shed light on this choice. Opinion 28/2024 of the European Data Protection Board, “on certain data protection aspects related to the processing of personal data in the context of AI models”, adopted on 17 December 2024 (its French version bears the note that it has not yet been proofread), recalls in its summary that there is no hierarchy between the legal bases and that it is for the controller to identify the appropriate one. The CNIL page “Development of AI systems: the CNIL’s recommendations for complying with the GDPR”, dated 22 July 2025, covers the development phase.
Information to individuals in Articles 13, 14 and 15
Article 13 covers data collected from the person: they receive in particular the purposes of the processing and its legal basis (paragraph 1, point (c)). Article 14 covers data obtained otherwise: the information also states their source (paragraph 2, point (f)). It provides an exception where the provision of this information “proves impossible or would involve a disproportionate effort” (paragraph 5, point (b)); appropriate measures are then required, “including making the information publicly available”.
These two articles, like Article 15 on the right of access, require mention of “the existence of automated decision-making, including profiling, referred to in Article 22(1) and (4)” and, at least in those cases, “meaningful information about the logic involved”.
What remains to be assessed is the level of detail of “meaningful information” about the logic of a model, and the threshold of “disproportionate” effort.
Article 22 and the decision taken without human intervention
“The data subject shall have the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her.” Paragraph 2 sets this right aside in three cases: a contract that makes the decision necessary, authorisation by Union or Member State law, the person’s explicit consent. In the first and third, paragraph 3 requires “at least” the right to obtain human intervention, to express one’s point of view and to contest the decision.
The article defines neither “based solely” nor “similarly significantly”: it does not set the share of human intervention beyond which a decision ceases to be solely automated. The project therefore documents what the system produces, what the person sees and what they can set aside.
Data protection by design and impact assessment
Article 25(1) calls for appropriate technical and organisational measures, “such as pseudonymisation”, implemented “both at the time of the determination of the means for processing and at the time of the processing itself”. In our reading, the architecture choices of an AI project fall within these means: columns kept for training, content of logs, retention period. Our article on data anonymisation deals with pseudonymisation.
Article 35(1) requires an impact assessment before the processing where it is “likely to result in a high risk to the rights and freedoms of natural persons”. Paragraph 3, point (a), requires one for the systematic and extensive evaluation of personal aspects by automated processing, where it forms the basis for decisions producing legal effects or significantly affecting the person. Determining whether a project presents a “high risk” is for the controller, who seeks the advice of its data protection officer (paragraph 2).
GDPR and artificial intelligence after Regulation (EU) 2024/1689
The AI Regulation leaves these obligations in place. Under its Article 2(7), read in the consolidated version of 27 July 2026, Union law on data protection applies to personal data processed in connection with the rights and obligations it lays down and, without prejudice to its Articles 4a and 59, this Regulation “shall not affect Regulation (EU) 2016/679”. Article 4a governs the exceptional processing of special categories of data to detect and correct bias, under conditions that are added to those of the GDPR.
The Commission proposal COM(2025) 837 final, of 19 November 2025, provides in particular for inserting into the GDPR an Article 88c on processing “in the context of the development and operation of AI” and for rewriting the first two paragraphs of Article 22. On the date of consultation, the Cellar repository offers only one consolidated version of the GDPR, dated 4 May 2016.
The alignment between the register of AI systems and the record of processing activities is described on our AI governance page; roles on data fall under data governance.
A project’s record sheet can carry, against each article read here, the answer adopted, its author and its date, before being handed to the data protection officer.