Personal data · Artificial intelligence
Your data to train AI: will you still be able to say no?
Discover this article in audio or video
Europe is debating the conditions for using personal data. A proposal does not, by itself, remove existing rights.
Case statusSources revisited on 6 October 2026 · general information, subject to the particular circumstances.
01
What is being discussed in Brussels
On 21 September, noyb criticised a European draft addressing, among other matters, personal data and AI. The organisation published a Council Presidency compromise draft dated 3 September 2026. Its statement is an advocacy position on negotiations, not a court ruling or evidence that a reform has become applicable. The debate must be distinguished from the law currently governing a particular use. [2,3]
The practical question deserves attention now: under what conditions could a business reuse personal data for AI development, and how could an individual challenge it? The final wording, adoption and timetable must be checked before announcing a change in rights. A negotiating document does not settle those questions. [3]
02
Objection does not work identically in every case
The GDPR remains the starting point. Where processing relies on legitimate interests or a public-interest task, Article 21 provides for objection on grounds relating to the individual’s particular situation. The controller must stop unless it demonstrates the compelling legitimate grounds specified in the provision or needs relating to legal claims. Objection to direct marketing follows a separate rule, without that balancing exercise. [1]
This means neither that every training operation requires consent nor that legitimate interests automatically authorise any collection. Purpose, necessity and the interests involved require examination; special-category data raises additional conditions. In Opinion 28/2024, the European Data Protection Board stresses a case-specific assessment of models and processing. Publicly available data does not become unrestricted material for reuse merely because it can be read online. [1,4]
Erasure and objection are also different requests. Deleting an account does not prove that previous use has been reversed or a model retrained. A useful response explains what is processed, on what basis and what the request changes. Avoid treating an unverified technical assurance as proof that a particular use has ended. [1]
03
The Law Right approach: keep the request and the response
- Identify the relevant notice: controller, stated purpose, categories of data, legal basis and rights channel. Keep a dated copy. Do not supply new sensitive documents to explain an objection unless necessary. A targeted request helps establish the issue without needlessly expanding the information disclosed.
- State precisely what you are asking and retain the acknowledgement. Where Article 21 is relevant, explain the elements of your particular situation rather than assuming the processing is direct marketing. Have an incomplete response or refusal assessed against the actual processing and the competent authority’s remit.
- For businesses, map the information supplied to AI tools and the uses permitted by the contracts. A reform under discussion does not replace assessment of present compliance. The next step is a targeted check, not an announcement that all protection has vanished. Keep contractual promises separate from what the service actually does.
Your data is not a detail.
Which data, which actor, which use? Start by making the processing visible.
Explanatory visualisation, not a measurement or an individual legal conclusion.Before deciding yes or no, identify the processing, its legal basis and the right you are exercising.
Sources and verificationOpen the sources +
- primaryRèglement (UE) 2016/679 · articles 5, 6, 9, 12–14, 17 et 21 · texte législatif ↗
- secondarynoyb · 21.09.2026 · position critique sur un projet, non droit adopté ↗
- primaryProjet de compromis de la présidence du Conseil · 03.09.2026 · 12535/26 · copie publiée par noyb, non droit adopté ↗
- primaryEDPB · Opinion 28/2024 · AI models and personal data · institutional interpretation ↗
