Wisdom is Knowing What not to Say: Hallucination-Free LLMs Unlearning via Attention Shifting

• URL: http://arxiv.org/abs/2510.17210v2
• Date: Sat, 01 Nov 2025 00:49:29 GMT
• Title: Wisdom is Knowing What not to Say: Hallucination-Free LLMs Unlearning via Attention Shifting
• Authors: Chenchen Tan, Youyang Qu, Xinghao Li, Hui Zhang, Shujie Cui, Cunjian Chen, Longxiang Gao,
• Abstract summary: We introduce Attention-Shifting (AS) framework for selective unlearning.<n>AS is driven by two design objectives: (1) context-preserving suppression that attenuates attention to fact-bearing tokens without disrupting LLMs' linguistic structure; and (2) hallucination-resistant response shaping that discourages fabricated completions when queried about unlearning content.<n> Experimental results show that AS improves performance over the state-of-the-art unlearning methods, achieving up to 15% higher accuracy on the ToFU benchmark and 10% on the TDEC benchmark, while maintaining competitive hallucination-free unlearning effectiveness.
• Score: 11.725875396424927
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The increase in computing power and the necessity of AI-assisted decision-making boost the growing application of large language models (LLMs). Along with this, the potential retention of sensitive data of LLMs has spurred increasing research into machine unlearning. However, existing unlearning approaches face a critical dilemma: Aggressive unlearning compromises model utility, while conservative strategies preserve utility but risk hallucinated responses. This significantly limits LLMs' reliability in knowledge-intensive applications. To address this, we introduce a novel Attention-Shifting (AS) framework for selective unlearning. AS is driven by two design objectives: (1) context-preserving suppression that attenuates attention to fact-bearing tokens without disrupting LLMs' linguistic structure; and (2) hallucination-resistant response shaping that discourages fabricated completions when queried about unlearning content. AS realizes these objectives through two attention-level interventions, which are importance-aware suppression applied to the unlearning set to reduce reliance on memorized knowledge and attention-guided retention enhancement that reinforces attention toward semantically essential tokens in the retained dataset to mitigate unintended degradation. These two components are jointly optimized via a dual-loss objective, which forms a soft boundary that localizes unlearning while preserving unrelated knowledge under representation superposition. Experimental results show that AS improves performance preservation over the state-of-the-art unlearning methods, achieving up to 15% higher accuracy on the ToFU benchmark and 10% on the TDEC benchmark, while maintaining competitive hallucination-free unlearning effectiveness. Compared to existing methods, AS demonstrates a superior balance between unlearning effectiveness, generalization, and response reliability.

Related papers

• Attention Smoothing Is All You Need For Unlearning [12.239021292288967]
  Large Language Models are prone to memorizing sensitive, copyrighted, or hazardous content, posing significant privacy and legal concerns.<n>We propose Attention Smoothing Unlearning, a principled framework that casts unlearning as self-distillation from a forget-teacher derived from the model's own attention.
  arXiv  Detail & Related papers  (2026-03-01T21:39:35Z)
• MeGU: Machine-Guided Unlearning with Target Feature Disentanglement [73.49657372882082]
  We propose a novel framework that guides unlearning through concept-aware re-alignment.<n>MeGU enables controlled and selective forgetting, effectively mitigating both under-unlearning and over-unlearning.
  arXiv  Detail & Related papers  (2026-02-19T05:20:31Z)
• AUVIC: Adversarial Unlearning of Visual Concepts for Multi-modal Large Language Models [63.05306474002547]
  Regulatory frameworks mandating the 'right to be forgotten' drive the need for machine unlearning.<n>We introduce AUVIC, a novel visual concept unlearning framework for MLLMs.<n>We show that AUVIC achieves state-of-the-art target forgetting rates while incurs minimal performance degradation on non-target concepts.
  arXiv  Detail & Related papers  (2025-11-14T13:35:32Z)
• MLLMEraser: Achieving Test-Time Unlearning in Multimodal Large Language Models through Activation Steering [36.80441487363007]
  MLLMEraser is an input-aware, training-free framework for test-time unlearning.<n>We construct a multimodal erasure direction by contrasting adversarially perturbed, knowledge-recall image-text pairs.<n>Experiments on LLaVA-1.5 and Qwen-2.5-VL demonstrate that MLLMEraser consistently outperforms state-of-the-art MLLM unlearning baselines.
  arXiv  Detail & Related papers  (2025-10-05T14:20:17Z)
• Understanding the Dilemma of Unlearning for Large Language Models [50.54260066313032]
  Unlearning seeks to remove specific knowledge from large language models (LLMs)<n>We propose unPact, an interpretable framework for unlearning via prompt attribution and contribution tracking.
  arXiv  Detail & Related papers  (2025-09-29T12:15:19Z)
• GRAIT: Gradient-Driven Refusal-Aware Instruction Tuning for Effective Hallucination Mitigation [62.63014905981601]
  Refusal-Aware Instruction Tuning (RAIT) aims to enhance Large Language Models (LLMs) by improving their ability to refuse responses to questions beyond their knowledge.<n>Effective RAIT must address two key challenges: firstly, effectively reject unknown questions to minimize hallucinations; secondly, avoid over-refusal to ensure questions that can be correctly answered are not rejected.<n>GraIT employs gradient-driven sample selection to effectively minimize hallucinations and (2) introduces an adaptive weighting mechanism during fine-tuning to reduce the risk of over-refusal.
  arXiv  Detail & Related papers  (2025-02-09T14:11:30Z)
• When Machine Unlearning Meets Retrieval-Augmented Generation (RAG): Keep Secret or Forget Knowledge? [27.718577066105485]
  Large language models (LLMs) can inadvertently learn and retain sensitive information and harmful content during training.<n>We propose a lightweight behavioral unlearning framework based on Retrieval-Augmented Generation (RAG) technology.<n>We evaluate our framework through extensive experiments on both open-source and closed-source models, including ChatGPT, Gemini, Llama-2-7b-chat, and PaLM 2.
  arXiv  Detail & Related papers  (2024-10-20T03:51:01Z)
• Towards Robust Knowledge Unlearning: An Adversarial Framework for Assessing and Improving Unlearning Robustness in Large Language Models [19.015202590038996]
  We design Dynamic Unlearning Attack (DUA), a dynamic and automated framework to attack unlearned models. We propose Latent Adrial Unlearning (LAU), a universal framework that effectively enhances the robustness of the unlearned process. We demonstrate that LAU improves unlearning effectiveness by over $53.5%$, cause only less than a $11.6%$ reduction in neighboring knowledge, and have almost no impact on the model's general capabilities.
  arXiv  Detail & Related papers  (2024-08-20T09:36:04Z)
• Towards Effective Evaluations and Comparisons for LLM Unlearning Methods [97.2995389188179]
  This paper seeks to refine the evaluation of machine unlearning for large language models.<n>It addresses two key challenges -- the robustness of evaluation metrics and the trade-offs between competing goals.
  arXiv  Detail & Related papers  (2024-06-13T14:41:00Z)
• Rethinking Machine Unlearning for Large Language Models [85.92660644100582]
  We explore machine unlearning in the domain of large language models (LLMs)<n>This initiative aims to eliminate undesirable data influence (e.g., sensitive or illegal information) and the associated model capabilities.
  arXiv  Detail & Related papers  (2024-02-13T20:51:58Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.