LLM Unlearning

LLM: Large Language Model

Large language model unlearning aims to remove targeted information, such as private data, hazardous knowledge, or memorized content, without retraining the model from scratch or damaging useful abilities. Common approaches use gradient-based fine-tuning with retention constraints, parameter or representation edits, and inference-time steering; retrieval-augmented systems can also forget by changing their external knowledge stores. Benchmarks such as TOFU, MUSE, and WMDP test forgetting alongside retained utility, while newer evaluations probe recovery through paraphrases, multi-hop reasoning, quantization, and interaction. A central challenge is that suppressing an answer does not necessarily erase the information internally, and stronger forgetting can cause collateral damage or remain reversible.

Latest papers

192 papers32 papers in the last four weeks, up 68% on the four weeks before. 0.2% of all new papers.

Preserving Utility in Continual LLM Unlearning via Orthogonal Projections
Oct 29, 2026Juan Belieni, Ana Carolina Erthal, Eliezer de Souza da Silva +1LLM Unlearning
School of Applied Mathematics, Getulio Vargas Foundation · University of Coimbra, CISUC/LASI, DEI
TRACE: Data-Free Text Reconstruction Attacks against Approximate Unlearning in LLMs
Oct 22, 2026Mengying Zhang, Derui Wang, Ibrahim Khalil +3Privacy Leakage in Language Models Training Data Reconstruction Attacks
RMIT University · CSIRO · The Hong Kong University of Science and Technology +1
Orthogonal Boundary Unlearning for Dual-Use Concepts in Large Language Models
Oct 20, 2026Niccolò RonchettiLLM Unlearning Machine Unlearning
Does Unlearning Survive Collaboration? Robust Unlearning for Shared-Model LLM Teams
Oct 19, 2026Hengsong Liu, Liangyu Teng, Jingnan Cai +5LLM Unlearning Machine Unlearning Evaluation
Fudan Institute on Networking Systems of AI
CoLU: Confidence-Weighted Layerwise Unlearning for Large Language Models
Oct 15, 2026Jingkai Guo, Ziyi Zhao, Chaitali Chakrabarti +1Machine Unlearning LLM Unlearning
School of Electrical, Computer and Energy Engineering, Arizona State University, Tempe, AZ, USA
Behavioral Forgetting Is Not Privacy in Machine Unlearning
Oct 15, 2026Tyler Lizzo, Larry HeckMachine Unlearning LLM Unlearning
AI Virtual Assistant (AVA) Lab, Georgia Institute of Technology
Bias Mitigation in Audio-Enabled LLMs through Adversarial Training and Distillation Methods
Oct 15, 2026Adnan Al Ali, Jindřich Libovický, Anne LauscherAdversarial Debiasing Language Model Debiasing
Charles University, Faculty of Mathematics and Physics, Czech Republic · Trustworthy AI Lab, University of Hamburg, Germany
FIRM: Fisher-Informed Retention Merging for Federated LLM Unlearning
Oct 15, 2026Reeyansh Shah, Maram Hasan, Subhasis Chaudhuri +1Federated Unlearning Machine Unlearning
Indian Institute of Technology Bombay
MemCut: Every Secret Leaves a Trail in the Long-Term Memory of LLM Agents
Oct 15, 2026Cheng Zhang, Xuanning Cao, Zhicheng Zheng +1LLM Agent Memory Privacy Leakage in Language Models
The Hong Kong University of Science and Technology (Guangzhou) · University of Science and Technology of China · The Hong Kong University of Science and Technology
Not All Change Is Damage: Update-Matched Unlearning for Large Language Models
Oct 15, 2026Jie Li, Youming Tao, Bowei He +1Machine Unlearning LLM Unlearning
University of Massachusetts Boston · Technische Universität Berlin · Mohamed bin Zayed University of Artificial Intelligence
RA $^2$ : Retain-Anchored Attraction Preserves Forget-Adjacent Utility in LLM Unlearning
Oct 14, 2026Zinan Ling, Yang Xiao, Ruimeng Ye +2Machine Unlearning LLM Unlearning
The University of Tulsa · Michigan State University
Can Agents Forget When Asked? A Benchmark for Memory Unlearning in LLM Agents
Oct 13, 2026Bofan Xie, Chaokun Wang, Guanchen Ge +4LLM Agent Memory Memory Agent Benchmarks
Tsinghua University · Sina Weibo
Can Agents Truly Forget? Explore Machine Unlearning for LLM Agents
Oct 13, 2026Junkai Lin, Ge Xu, Zheng Xu +2Machine Unlearning LLM Unlearning
Institute of Automation, Chinese Academy of Sciences · The Chinese University of Hong Kong · Nanyang Technological University +2
Decouple to Forget: Semantic Mediation for Sensitive Unlearning in Large Language Models
Oct 13, 2026Jingyue Cong, Andy Song, Yang Yu +7LLM Unlearning Privacy Leakage in Language Models
RMIT University, Melbourne, Australia · Yep AI, Melbourne, Australia · Monash University, Melbourne, Australia +1
REDaCT: Collapse-Free Unlearning in Large Reasoning Models
Oct 8, 2026Joshua Maccoby, Wendy Hui WangMachine Unlearning LLM Unlearning
Stevens Institute of Technology Hoboken, New Jersey, USA
Not Every Change Is Necessary: Recoverable Drift in Large Language Model Unlearning
Oct 8, 2026Xunlei Chen, Qinghui Gong, Jingkun Xue +3Machine Unlearning LLM Unlearning
University of Electronic Science and Technology of China · Southwest Jiaotong University
Nullify: Null-Space Activation Steering for Training-Free LLM Unlearning
Oct 7, 2026Wei Zhai, Xiang Liu, Qiang Huang +6LLM Unlearning Privacy Leakage in Language Models
College of Computer Science and Artificial Intelligence, Fudan University · BEDI Cloud · School of Electrical and Information Engineering, Tianjin University +3
How Learning Governs Unlearning across the Memorization-Generalization Spectrum
Oct 6, 2026Hwiyeong Lee, Hyelim Lim, Ingyu Bang +2Memorization in Language Models Machine Unlearning
Seoul National University · Hanyang University · Chung-Ang University
Linguistic Loopholes in LLM Unlearning: From a 174-Language Benchmark to Coverage-Aware Unlearning
Sep 30, 2026Tyler Skow, Shravan Chaudhari, Rama Chellappa +1Machine Unlearning LLM Unlearning
Johns Hopkins University
LLM Persona Unlearning
Sep 30, 2026Kemou Li, Zhuan Shi, Qizhou Wang +4Model Editing Persona-Conditioned Generation
State Key Laboratory of Internet of Things for Smart City, University of Macau · Mila – Québec AI Institute · McGill University +3
Preemptive LLM Unlearning against Forbidden Capability Acquisition via Gradient Sealing
Sep 30, 2026Kemou Li, Qizhou Wang, Yue Wang +6LLM Security LLM Safety
State Key Laboratory of Internet of Things for Smart City, University of Macau · RIKEN Center for Advanced Intelligence Project · The University of Melbourne +5
Faithful Dual-constrained Erasure for Robust LLM Safety Alignment
Sep 30, 2026Jiaqing Li, Shide Zhou, Zhibo Zhang +3LLM Safety Alignment Machine Unlearning
Huazhong University of Science and Technology · Hubei University
Learning What to Forget: Distributional Unlearning for LLM Representation Spaces
Sep 30, 2026Pinaki Mohanty, Haoran Tang, Maggie Makar +1Density Ratio Estimation Machine Unlearning
Department of Computer Science, College of Science & College of Engineering, Purdue University, West Lafayette, IN, USA · Computer Science and Engineering Division, University of Michigan, Ann Arbor, MI, USA
Unlearning Deceptive Behaviors in LLMs with Contrastive Forget Sets
Sep 30, 2026Haoran Tang, Rajiv KhannaDeception in Language Models Machine Unlearning
Department of Computer Science Purdue University
Storage Is Not Strategy: State-Conditioned Support Control for LLM Unlearning
Sep 29, 2026Tianhao Qian, Ziming Hong, Chongyang Gao +2Machine Unlearning LLM Unlearning
Southeast University (seu.edu.cn) · University of Sydney (usyd.edu.au) · Northwestern University (northwestern.edu) +2
LLM unbranding: Erasing Commercial Identity while Preserving Generic Utility
Sep 29, 2026Kajetan Ożóg, Alicja Wojciechowska, Dawid Malarz +3Prompt Optimization LLM Unlearning
Jagiellonian University · IDEAS Research Institute · Heinrich Heine Universität Düsseldorf
UnlearningSoup: Is Repeated Tuning Necessary for Large Language Model Unlearning?
Sep 29, 2026Puning Yang, Qizhou Wang, Junchi Yu +2LLM Unlearning Hyperparameter Optimization
Division of Computing and Mathematical Sciences, MBZUAI · RIKEN Center for Advanced Intelligence Project · TMLR Group, Department of Computer Science, Hong Kong Baptist University +1
Do LLMs Really Forget? Hidden-State Leakage in Model Unlearning and How to Fix it
Sep 29, 2026Hadi Reisizadeh, Jiajun Ruan, Sijia Liu +1Machine Unlearning LLM Unlearning
University of Minnesota · Michigan State University · IBM Research
Eternal Sunshine of the Spotless Mind: Systematically Erasing LLM's Memories
Sep 29, 2026Olga OhrimenkoLLM Unlearning Persistent Memory for Language Models
School of Computing and Information Systems The University of Melbourne
UNBIND: UNlearning By INference-time Directional Steering for Code LLMs
Sep 28, 2026Zhengyang Shan, Jiayun Xin, Yanjun Lin +7Activation Steering LLM Unlearning
Department of Computer Science and Technology Shandong University Qingdao, China

Oct 29, 2026J/K move · Enter open · S save

Preserving Utility in Continual LLM Unlearning via Orthogonal Projections

Juan Belieni, Ana Carolina Erthal, Eliezer de Souza da Silva, Diego Mesquita

School of Applied Mathematics, Getulio Vargas Foundation · University of Coimbra, CISUC/LASI, DEI

LLM Unlearning

Open paper AskPDF arXiv

Unlearning concerns the task of removing the influence of a sensitive portion of the training data, such as protected, copyrighted or private text, from a model without retraining it from scratch. While unlearning requests happen continuously during a model's lifetime, the vast majority of LLM unlearning methods are ill-suited for iterated removal, yielding steep decay in forgetting and utility upon sequential unlearning. This decay stems from interference between successive updates, as each new deletion can undo earlier ones or push the model further along directions already degraded by them. To address this, we introduce Continual Orthogonal Projection (COP), a gradient-projection scheme that restricts each update to the orthogonal complement of activation subspaces cached from prior forget sets. This approach provably guarantees first-order non-interference with earlier deletions. COP stores only a low-rank basis per layer, requiring neither past gradients nor previous forget sets, and plugs into unlearning methods that update the model through gradient steps on a differentiable loss without modifying its objective. Experiments on standard LLM unlearning benchmarks (TOFU and RWKU) show that COP preserves utility across sequential requests without weakening forgetting, preventing the collapse of standard objectives and forgetting more effectively than continual-specific baselines.