Recent unlearning methods for LLMs are vulnerable to relearning attacks: knowledge believed-to-be-unlearned re-emerges by fine-tuning on a small set of (even seemingly-unrelated) examples. We study this phenomenon in a controlled setting for example-level unlearning in vision classifiers. We make the surprising discovery that forget-set accuracy can recover from around 50% post-unlearning to nearly 100% with fine-tuning on just the retain set -- i.e., zero examples of the forget set. We observe this effect across a wide variety of unlearning methods, whereas for a model retrained from scratch excluding the forget set (gold standard), the accuracy remains at 50%. We observe that resistance to relearning attacks can be predicted by weight-space properties, specifically, L2-distance and linear mode connectivity between the original and the unlearned model. Leveraging this insight, we propose a new class of methods that achieve state-of-the-art resistance to relearning attacks.
@article{arxiv.2505.22310,
title = {From Dormant to Deleted: Tamper-Resistant Unlearning Through Weight-Space Regularization},
author = {Shoaib Ahmed Siddiqui and Adrian Weller and David Krueger and Gintare Karolina Dziugaite and Michael Curtis Mozer and Eleni Triantafillou},
journal= {arXiv preprint arXiv:2505.22310},
year = {2026}
}