# Unlearning Is Not Just Erasing: Temporal Decoupling via Generation Inequality

arXiv: https://arxiv.org/abs/2608.23020 · HTML: https://mlsanity.com/paper/2608.23020/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Machine unlearning, one of 12 directions in LLM safety and security.
- Builds on 21 of its 37 references that are in the map.
- First posted on arXiv 24 Aug 2026.

## Placed in

- LLM safety and security / [Machine unlearning](https://mlsanity.com/safety/machine-unlearning/)

## Builds on (most cited first, 21 of 21)

- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [BioBERT: a pre-trained biomedical language representation model for biomedical text mining](https://mlsanity.com/paper/1901.08746/) (2019)
- [TOFU: A Task of Fictitious Unlearning for LLMs](https://mlsanity.com/paper/2401.06121/) (2024)
- [The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning](https://mlsanity.com/paper/2403.03218/) (2024)
- [Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning](https://mlsanity.com/paper/2404.05868/) (2024)
- [Unrolling SGD: Understanding Factors Influencing Machine Unlearning](https://mlsanity.com/paper/2109.13398/) (2021)
- [MUSE: Machine Unlearning Six-Way Evaluation for Language Models](https://mlsanity.com/paper/2407.06460/) (2024)
- [Critical Tokens Matter: Token-Level Contrastive Estimation Enhances LLM's Reasoning Capability](https://mlsanity.com/paper/2411.19943/) (2024)
- [LLM Unlearning via Loss Adjustment with Only Forget Data](https://mlsanity.com/paper/2410.11143/) (2024)
- [Learning to Unlearn: Instance-wise Unlearning for Pre-trained Classifiers](https://mlsanity.com/paper/2301.11578/) (2023)
- [Large Language Model Safety: A Holistic Survey](https://mlsanity.com/paper/2412.17686/) (2024)
- [Selective Forgetting: Advancing Machine Unlearning Techniques and Evaluation in Language Models](https://mlsanity.com/paper/2402.05813/) (2024)
- [Unlearning Backdoor Attacks for LLMs with Weak-to-Strong Knowledge Distillation](https://mlsanity.com/paper/2410.14425/) (2024)
- [CLIPErase: Efficient Unlearning of Visual-Textual Associations in CLIP](https://mlsanity.com/paper/2410.23330/) (2024)
- [Exact and Efficient Unlearning for Large Language Model-based Recommendation](https://mlsanity.com/paper/2404.10327/) (2024)
- [SEUF: Is Unlearning One Expert Enough for Mixture-of-Experts LLMs?](https://mlsanity.com/paper/2411.18797/) (2024)
- [Agents Are All You Need for LLM Unlearning](https://mlsanity.com/paper/2502.00406/) (2025)
- [Wisdom is Knowing What not to Say: Hallucination-Free LLMs Unlearning via Attention Shifting](https://mlsanity.com/paper/2510.17210/) (2025)
- [CAP: Controllable Alignment Prompting for Unlearning in LLMs](https://mlsanity.com/paper/2604.21251/) (2026)
- [ALTER: Asymmetric LoRA for Token-Entropy-Guided Unlearning of LLMs](https://mlsanity.com/paper/2603.01792/) (2026)
- [The Unlearning Mirage: A Dynamic Framework for Evaluating LLM Unlearning](https://mlsanity.com/paper/2603.11266/) (2026)

## Abstract (arXiv)

Large language models (LLMs) require effective unlearning to address privacy regulations and safety concerns. However, achieving precise forgetting without compromising general utility remains challenging. Existing sequence- and token-level methods penalize target outputs without modeling their context-dependent retrieval paths, which can disrupt linguistic structure or suppress benign knowledge. We present ADU, a fine-grained, training-based framework that shifts unlearning from token erasure to contextual attention-pathway decoupling. Exploiting the functional distinction between local and global attention heads, ADU identifies preplan positions that retrieve persistent sensitive anchors and fixes their candidate paths under the original model. It then trains attention-projection adapters to suppress attention mass along these paths while preserving local-attention structure and retain-set language modeling. Post-training activation exchange tests whether the modified attention-output module transmits the learned forgetting effect. ADU achieves the strongest aggregate performance among evaluated baselines on the TOFU and WMDP benchmarks, including a Forget Quality of (0.93) on TOFU. It preserves 87--98% of model utility (92.9% on average versus 81.9% for baselines) while reducing side effects in benign contexts.