← Retour aux analyses
Cet article n'est pas encore disponible dans votre langue ; la version anglaise est affichée.
IA et Droit
The right to erasure meets large models: the technical puzzle of deletion
A user asks to delete their data, but it may already be baked into model weights — responding to erasure is a new problem.
Data-protection law grants a right to erasure, but once personal data is used in training, its influence may be fixed in model weights; deleting a database record does not truly 'delete' it.
This raises technical questions like machine unlearning and the legal difficulty of what counts as 'deleted'.
The pragmatic approach is to minimize baking identifiable personal data into models, keep a data-to-model mapping, and build an accountable process and explanation for deletion requests.