arXiv:2605.19042v1 Announce Type: new Abstract: Machine unlearning aims to remove the contribution of designated training data from a trained model while preserving performance on the remaining data. Existing work mainly focuses on single-task settings, whereas modern models ofte