mlsanityData to 13 Oct 2026

MUSE: Machine Unlearning Six-Way Evaluation for Language Models

arXiv 2407.06460 ↗First posted 8 Jul 2024cs.CL cs.AIcited by 257

On the map: LLM safety and security → Machine unlearning

Abstract (arXiv)

Language models (LMs) are trained on vast amounts of text data, which may include private and copyrighted content. Data owners may request the removal of their data from a trained model due to privacy or copyright concerns. However, exactly unlearning only these datapoints (i.e., retraining with the data removed) is intractable in modern-day models. This has led to the development of many approximate unlearning algorithms. The evaluation of the efficacy of these algorithms has traditionally been narrow in scope, failing to precisely quantify the success and practicality of the algorithm from the perspectives of both the model deployers and the data owners. We address this issue by proposing MUSE, a comprehensive machine unlearning evaluation benchmark that enumerates six diverse desirable properties for unlearned models: (1) no verbatim memorization, (2) no knowledge memorization, (3) no privacy leakage, (4) utility preservation on data not intended for removal, (5) scalability with respect to the size of removal requests, and (6) sustainability over sequential unlearning requests. Using these criteria, we benchmark how effectively eight popular unlearning algorithms on 7B-parameter LMs can unlearn Harry Potter books and news articles. Our results demonstrate that most algorithms can prevent verbatim memorization and knowledge memorization to varying degrees, but only one algorithm does not lead to severe privacy leakage. Furthermore, existing algorithms fail to meet deployer's expectations because they often degrade general model utility and also cannot sustainably accommodate successive unlearning requests or large-scale content removal. Our findings identify key issues with the practicality of existing unlearning algorithms on language models, and we release our benchmark to facilitate further evaluations: muse-bench.github.io

Key facts. Cited by 257 papers in the map (data through 13 Oct 2026). Placed in Machine unlearning, one of 12 directions in LLM safety and security. Builds on 55 of its 79 references that are in the map. First posted on arXiv 8 Jul 2024.

Builds on

55 of 79 references.

Closest ancestors · 22

Same direction, newest first

  1. 2024UnUnlearning: Unlearning is not sufficient for content regulation in advanced generative AI27 Jun
  2. 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning8 Apr
  3. 2024Inexact Unlearning Needs More Careful Evaluations to Avoid a False Sense of Privacy2 Mar
  4. 2024Eight Methods to Evaluate Robust Unlearning in LLMs26 Feb
  5. 2024Rethinking Machine Unlearning for Large Language Models13 Feb
  6. 2024TOFU: A Task of Fictitious Unlearning for LLMs11 Jan
  7. 2023Knowledge Unlearning for LLMs: Tasks, Methods, and Challenges27 Nov
  8. 2023SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation19 Oct
  9. 2023Large Language Model Unlearning14 Oct
  10. 2023In-Context Unlearning: Language Models as Few Shot Unlearners11 Oct
  11. 2023Who's Harry Potter? Approximate Unlearning in LLMs3 Oct
  12. 2023Fair Machine Unlearning: Data Removal while Mitigating Disparities27 Jul
10 more
  1. 2023Ticketed Learning-Unlearning Schemes27 Jun
  2. 2023To Be Forgotten or To Be Fair: Unveiling Fairness Implications of Machine Unlearning Methods7 Feb
  3. 2022Forget Unlearning: Towards True Data-Deletion in Machine Learning17 Oct
  4. 2022Knowledge Unlearning for Mitigating Privacy Risks in Language Models4 Oct
  5. 2022A Survey of Machine Unlearning6 Sep
  6. 2022Quark: Controllable Text Generation with Reinforced Unlearning26 May
  7. 2021Unrolling SGD: Understanding Factors Influencing Machine Unlearning27 Sep
  8. 2021Adaptive Machine Unlearning8 Jun
  9. 2021Remember What You Want to Forget: Algorithms for Machine Unlearning4 Mar
  10. 2021Machine Unlearning via Algorithmic Stability25 Feb

Other references · 33

  1. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,986
  2. 2022Locating and Editing Factual Associations in GPTcited by 2,650
  3. 2016Membership Inference Attacks against Machine Learning Modelscited by 2,273
  4. 2020Extracting Training Data from Large Language Modelscited by 2,181
  5. 2019Machine Unlearningcited by 1,056
  6. 2019Eternal Sunshine of the Spotless Net: Selective Forgetting in Deep Networkscited by 617
  7. 2023Erasing Concepts from Diffusion Modelscited by 535
  8. 2019Certified Data Removal from Machine Learning Modelscited by 532
  9. 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearningcited by 498
  10. 2019Making AI Forget You: Data Deletion in Machine Learningcited by 462
  11. 2019Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classificationcited by 400
  12. 2023Detecting Pretraining Data from Large Language Modelscited by 383
21 more
  1. 2023Multi-step Jailbreaking Privacy Attacks on ChatGPTcited by 333
  2. 2020Descent-to-Delete: Gradient-Based Methods for Machine Unlearningcited by 310
  3. 2023Forget-Me-Not: Learning to Forget in Text-to-Image Diffusion Modelscited by 301
  4. 2023LEACE: Perfect linear concept erasure in closed formcited by 259
  5. 2020Approximate Data Deletion from Machine Learning Modelscited by 255
  6. 2021Enhanced Membership Inference Attacks against Machine Learning Modelscited by 247
  7. 2024Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modificationscited by 227
  8. 2020Information Leakage in Embedding Modelscited by 203
  9. 2020Forgetting Outside the Box: Scrubbing Deep Networks of Information Accessible from Input-Output Observationscited by 180
  10. 2020DeltaGrad: Rapid retraining of machine learning modelscited by 149
  11. 2023Foundation Models and Fair Usecited by 116
  12. 2022Federated Unlearning: How to Efficiently Erase a Client in FL?cited by 95
  13. 2023In-context Pretraining: Language Modeling Beyond Document Boundariescited by 78
  14. 2023DEPN: Detecting and Editing Privacy Neurons in Pretrained Language Modelscited by 77
  15. 2023SILO Language Models: Isolating Legal Risk In a Nonparametric Datastorecited by 71
  16. 2024Evaluating Copyright Takedown Methods for Language Modelscited by 43
  17. 2023Privacy Implications of Retrieval-Based Language Modelscited by 32
  18. 2022Learning with Recoverable Forgettingcited by 30
  19. 2020Federated Unlearningcited by 29
  20. 2022Making Recommender Systems Forget: Learning and Unlearning for Erasable Recommendationcited by 29
  21. 2024Fantastic Copyrighted Beasts and How (Not) to Generate Themcited by 24

Built on by

Cited by year (2026 to 13 Oct)

Citing papers per year: 2024: 20; 2025: 110; 2026: 127.

257 papers.

Sort:

By area:

  1. 2024Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearningcited by 167
  2. 2024An Adversarial Perspective on Machine Unlearning for AI Safetycited by 119
  3. 2025OpenUnlearning: Accelerating LLM Unlearning via Unified Benchmarking of Methods and Metricscited by 82
  4. 2024Catastrophic Failure of LLM Unlearning via Quantizationcited by 79
  5. 2025Towards LLM Unlearning Resilient to Relearning Attacks: A Sharpness-Aware Minimization Perspective and Beyondcited by 78
  6. 2024LLM Unlearning via Loss Adjustment with Only Forget Datacited by 74
  7. 2024Position: LLM Unlearning Benchmarks are Weak Measures of Progresscited by 62
  8. 2024A Closer Look at Machine Unlearning for Large Language Modelscited by 55
  9. 2025Stepwise Perplexity-Guided Refinement for Efficient Chain-of-Thought Reasoning in Large Language Modelscited by 51
  10. 2025Unlearning Isn't Deletion: Investigating Reversibility of Machine Unlearning in LLMscited by 41
  11. 2025Exploring Criteria of Loss Reweighting to Enhance LLM Unlearningcited by 39
  12. 2025When is Task Vector Provably Effective for Model Editing? A Generalization Analysis of Nonlinear Transformerscited by 38
38 more
  1. 2024CLEAR: Character Unlearning in Textual and Visual Modalitiescited by 36
  2. 2025Rethinking Memory in LLM based Agents: Representations, Operations, and Emerging Topicscited by 36
  3. 2024On Effects of Steering Latent Representation for Large Language Model Unlearningcited by 30
  4. 2024Unlearning as multi-task optimization: A normalized gradient difference approach with an adaptive learning ratecited by 30
  5. 2025Representation Bending for Large Language Model Safetycited by 29
  6. 2024Towards Robust Knowledge Unlearning: An Adversarial Framework for Assessing and Improving Unlearning Robustness in Large Language Modelscited by 27
  7. 2025BLUR: A Bi-Level Optimization Approach for LLM Unlearningcited by 23
  8. 2025LUME: LLM Unlearning with Multitask Evaluationscited by 22
  9. 2025ReLearn: Unlearning via Learning for Large Language Modelscited by 21
  10. 2025SAEs Can Improve Unlearning: Dynamic Sparse Autoencoder Guardrails for Precision Unlearning in LLMscited by 21
  11. 2025Invariance Makes LLM Unlearning Resilient Even to Unanticipated Downstream Fine-Tuningcited by 21
  12. 2024Unlearn and Burn: Adversarial Machine Unlearning Requests Destroy Model Accuracycited by 17
  13. 2025GUARD: Generation-time LLM Unlearning via Adaptive Restriction and Detectioncited by 17
  14. 2025LLM Unlearning with LLM Beliefscited by 17
  15. 2025Safety Mirage: How Spurious Correlations Undermine VLM Safety Fine-Tuning and Can Be Mitigated by Machine Unlearningcited by 16
  16. 2024From Transparency to Accountability and Back: A Discussion of Access and Evidence in AI Auditingcited by 15
  17. 2025Position: Adversarial ML for LLMs Is Not Making Any Progresscited by 15
  18. 2025FlexOlmo: Open Language Models for Flexible Data Usecited by 15
  19. 2024Evaluating Deep Unlearning in Large Language Modelscited by 14
  20. 2024SEUF: Is Unlearning One Expert Enough for Mixture-of-Experts LLMs?cited by 14
  21. 2025A General Framework to Enhance Fine-tuning-based LLM Unlearningcited by 14
  22. 2025Exact Unlearning of Finetuning Data via Model Merging at Scalecited by 14
  23. 2025Do LLMs Really Forget? Evaluating Unlearning with Knowledge Correlation and Confidence Awarenesscited by 14
  24. 2025BLUR: A Benchmark for LLM Unlearning Robust to Forget-Retain Overlapcited by 13
  25. 2025GRU: Mitigating the Trade-off between Unlearning and Retention for LLMscited by 12
  26. 2025LLM Unlearning Reveals a Stronger-Than-Expected Coreset Effect in Current Benchmarkscited by 11
  27. 2025Towards Lifecycle Unlearning Commitment Management: Measuring Sample-level Unlearning Completenesscited by 11
  28. 2024The Future of Open Human Feedbackcited by 10
  29. 2025SoK: Machine Unlearning for Large Language Modelscited by 10
  30. 2025Unlearning Isn't Invisible: Detecting Unlearning Traces in LLMs from Model Outputscited by 10
  31. 2025Unlearning That Lasts: Utility-Preserving, Robust, and Almost Irreversible Forgetting in LLMscited by 10
  32. 2026Maximizing Local Entropy Where It Matters: Prefix-Aware Localized LLM Unlearningcited by 10
  33. 2024Underestimated Privacy Risks for Minority Populations in Large Language Model Unlearningcited by 9
  34. 2025Keeping an Eye on LLM Unlearning: The Hidden Risk and Remedycited by 9
  35. 2025UCD: Unlearning in LLMs via Contrastive Decodingcited by 9
  36. 2025Rectifying Privacy and Efficacy Measurements in Machine Unlearning: A New Inference Attack Perspectivecited by 9
  37. 2025Leak@k: Unlearning Does Not Make LLMs Forget Under Probabilistic Decodingcited by 9
  38. 2025Reviving Your MNEME: Predicting The Side Effects of LLM Unlearning and Fine-Tuning via Sparse Model Diffingcited by 8

50 most cited of 257 All 257 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024Towards Effective Evaluations and Comparisons for LLM Unlearning Methods
  2. 2025A Comprehensive Survey of Machine Unlearning Techniques for Large Language Models
  3. 2026Unlearning in LLMs: Methods, Evaluation, and Open Challenges
  4. 2024The Frontier of Data Erasure: Machine Unlearning for Large Language Models
  5. 2026Open-MMUnlearning: Unifying Methods and Evaluation for MLLM Unlearning
  6. 2024Machine Unlearning in Large Language Models
  7. 2025OBLIVIATE: Robust and Practical Machine Unlearning for Large Language Models
  8. 2024Machine Unlearning of Pre-trained Large Language Models