Attention Smoothing Is All You Need For Unlearning

ICLR 2026 Conference Submission9688 Authors

Keywords: Large Language Model, Large Language Model Unlearning, Self-distillation, Attention Smoothing

TL;DR: We unlearn by smoothing attention: raise the softmax temperature to form a forget-teacher and distill the model, removing targeted facts while keeping outputs on forget prompts coherent, with minimal degradation in utility.

Abstract: Large Language Models are prone to memorizing sensitive, copyrighted, or hazardous content, posing significant privacy and legal concerns. Retraining from scratch is computationally infeasible, whereas current unlearning methods exhibit unstable trade-offs between forgetting and utility, frequently producing incoherent outputs on forget prompts and failing to generalize due to the persistence of lexical-level and semantic-level associations in attention. We propose Attention Smoothing Unlearning (ASU), a principled framework that casts unlearning as self-distillation from a forget-teacher derived from the model’s own attention. By increasing the softmax temperature, ASU flattens attention distributions and directly suppresses the lexical-level and semantic-level associations responsible for reconstructing memorized knowledge. This results in a bounded optimization objective that erases factual information yet maintains coherence in responses to forget prompts. Empirical evaluation on TOFU, MUSE, and WMDP, along with real-world and continual unlearning scenarios across Question and Answer (QA) and text completion, demonstrates that ASU outperforms the baselines for most of the unlearning scenarios, delivering robust unlearning with minimal loss of model utility.

Primary Area: alignment, fairness, safety, privacy, and societal considerations

Submission Number: 9688