mlsanityData to 14 Oct 2026

Catastrophic Failure of LLM Unlearning via Quantization

arXiv 2410.16454 ↗First posted 21 Oct 2024cs.CL cs.AIcited by 81

On the map: LLM safety and security → Machine unlearning

Abstract (arXiv)

Large language models (LLMs) have shown remarkable proficiency in generating text, benefiting from extensive training on vast textual corpora. However, LLMs may also acquire unwanted behaviors from the diverse and sensitive nature of their training data, which can include copyrighted and private content. Machine unlearning has been introduced as a viable solution to remove the influence of such problematic content without the need for costly and time-consuming retraining. This process aims to erase specific knowledge from LLMs while preserving as much model utility as possible. Despite the effectiveness of current unlearning methods, little attention has been given to whether existing unlearning methods for LLMs truly achieve forgetting or merely hide the knowledge, which current unlearning benchmarks fail to detect. This paper reveals that applying quantization to models that have undergone unlearning can restore the "forgotten" information. To thoroughly evaluate this phenomenon, we conduct comprehensive experiments using various quantization techniques across multiple precision levels. We find that for unlearning methods with utility constraints, the unlearned model retains an average of 21% of the intended forgotten knowledge in full precision, which significantly increases to 83% after 4-bit quantization. ... Our code is available at: hrefhttps://github.com/zzwjames/FailureLLMUnlearninghttps://github.com/zzwjames/FailureLLMUnlearning.

Key facts. Cited by 81 papers in the map (data through 14 Oct 2026). Placed in Machine unlearning, one of 12 directions in LLM safety and security. Builds on 66 of its 78 references that are in the map. First posted on arXiv 21 Oct 2024.

Builds on

66 of 78 references.

Closest ancestors · 23

Same direction, newest first

  1. 2024Composable Interventions for Language Models9 Jul
  2. 2024MUSE: Machine Unlearning Six-Way Evaluation for Language Models8 Jul
  3. 2024RWKU: Benchmarking Real-World Knowledge Unlearning for Large Language Models16 Jun
  4. 2024To Each (Textual Sequence) Its Own: Improving Memorized-Data Unlearning in Large Language Models6 May
  5. 2024SOUL: Unlocking the Power of Second-Order Optimization for LLM Unlearning28 Apr
  6. 2024Offset Unlearning for Large Language Models17 Apr
  7. 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning8 Apr
  8. 2024Challenging Forgets: Unveiling the Worst-Case Forget Sets in Machine Unlearning12 Mar
  9. 2024Guardrail Baselines for Unlearning in LLMs5 Mar
  10. 2024Eight Methods to Evaluate Robust Unlearning in LLMs26 Feb
  11. 2024Towards Safer Large Language Models through Machine Unlearning15 Feb
  12. 2024Rethinking Machine Unlearning for Large Language Models13 Feb
11 more
  1. 2024TOFU: A Task of Fictitious Unlearning for LLMs11 Jan
  2. 2023Unlearn What You Want to Forget: Efficient Unlearning for LLMs31 Oct
  3. 2023SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation19 Oct
  4. 2023Large Language Model Unlearning14 Oct
  5. 2023In-Context Unlearning: Language Models as Few Shot Unlearners11 Oct
  6. 2023Who's Harry Potter? Approximate Unlearning in LLMs3 Oct
  7. 2023Machine Unlearning: Solutions and Challenges14 Aug
  8. 2022Knowledge Unlearning for Mitigating Privacy Risks in Language Models4 Oct
  9. 2022Continual Learning and Private Unlearning24 Mar
  10. 2021Unrolling SGD: Understanding Factors Influencing Machine Unlearning27 Sep
  11. 2021Remember What You Want to Forget: Algorithms for Machine Unlearning4 Mar

Other references · 43

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2023GPT-4 Technical Reportcited by 18,396
  3. 2024The Llama 3 Herd of Modelscited by 14,715
  4. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 13,028
  5. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  6. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  7. 2023QLoRA: Efficient Finetuning of Quantized LLMscited by 3,418
  8. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,975
  9. 2017TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehensioncited by 2,955
  10. 2022Locating and Editing Factual Associations in GPTcited by 2,691
  11. 2023A Survey of Large Language Modelscited by 2,576
  12. 2017Decoupled Weight Decay Regularizationcited by 1,831
31 more
  1. 2023Jailbreaking Black Box Large Language Models in Twenty Queriescited by 1,420
  2. 2023AWQ: Activation-aware Weight Quantization for LLM Compression and Accelerationcited by 1,298
  3. 2022Editing Models with Task Arithmeticcited by 1,120
  4. 2019Machine Unlearningcited by 1,061
  5. 2021Knowledge Neurons in Pretrained Transformerscited by 601
  6. 2019Certified Data Removal from Machine Learning Modelscited by 535
  7. 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearningcited by 509
  8. 2019Making AI Forget You: Data Deletion in Machine Learningcited by 465
  9. 2023Detecting Pretraining Data from Large Language Modelscited by 387
  10. 2023SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compressioncited by 367
  11. 2023SqueezeLLM: Dense-and-Sparse Quantizationcited by 294
  12. 2024TrustLLM: Trustworthiness in Large Language Modelscited by 285
  13. 2023A Survey on Model Compression for Large Language Modelscited by 269
  14. 2024Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modificationscited by 231
  15. 2022Are Large Pre-Trained Language Models Leaking Your Personal Information?cited by 200
  16. 2024Latent Adversarial Training Improves Robustness to Persistent Harmful Behaviors in LLMscited by 173
  17. 2023Can Sensitive Information Be Deleted From LLMs? Objectives for Defending Against Extraction Attackscited by 158
  18. 2024BiLLM: Pushing the Limit of Post-Training Quantization for LLMscited by 144
  19. 2018Generating Informative and Diverse Conversational Responses via Adversarial Information Maximizationcited by 116
  20. 2024On Protecting the Data Privacy of Large Language Models (LLMs): A Surveycited by 96
  21. 2023In-context Pretraining: Language Modeling Beyond Document Boundariescited by 79
  22. 2023DEPN: Detecting and Editing Privacy Neurons in Pretrained Language Modelscited by 79
  23. 2024OneBit: Towards Extremely Low-bit Large Language Modelscited by 79
  24. 2024Exploiting LLM Quantizationcited by 65
  25. 2024Decoding Compressed Trust: Scrutinizing the Trustworthiness of Efficient LLMs Under Compressioncited by 57
  26. 2024Accurate LoRA-Finetuning Quantization of LLMs via Information Retentioncited by 41
  27. 2024Digger: Detecting Copyright Content Mis-usage in Large Language Model Trainingcited by 25
  28. 2024Unlocking Memorization in Large Language Models with Dynamic Soft Promptingcited by 21
  29. 2024InfuserKI: Enhancing Large Language Models with Knowledge Graphs via Infuser-Guided Knowledge Integrationcited by 19
  30. 2024Selection of LLM Fine-Tuning Data based on Orthogonal Rulescited by 12
  31. 2024Evaluating the Generalization Ability of Quantized LLMs: Benchmark, Analysis, and Toolboxcited by 11

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 34; 2026: 47.

81 papers.

Sort:

By area:

  1. 2025The Attacker Moves Second: Stronger Adaptive Attacks Bypass Defenses Against Llm Jailbreaks and Prompt Injectionscited by 129
  2. 2025OpenUnlearning: Accelerating LLM Unlearning via Unified Benchmarking of Methods and Metricscited by 83
  3. 2025Unlearning Isn't Deletion: Investigating Reversibility of Machine Unlearning in LLMscited by 42
  4. 2025Model Tampering Attacks Enable More Rigorous Evaluations of LLM Capabilitiescited by 39
  5. 2025A Comprehensive Survey of Machine Unlearning Techniques for Large Language Modelscited by 34
  6. 2025AgentTTS: Large Language Model Agent for Test-time Compute-optimal Scaling Strategy in Complex Taskscited by 16
  7. 2025A General Framework to Enhance Fine-tuning-based LLM Unlearningcited by 14
  8. 2025SAUCE: Selective Concept Unlearning in Vision-Language Models with Sparse Autoencoderscited by 13
  9. 2026Shaping capabilities with token-level data filteringcited by 13
  10. 2025From Dormant to Deleted: Tamper-Resistant Unlearning Through Weight-Space Regularizationcited by 12
  11. 2025Distillation Robustifies Unlearningcited by 11
  12. 2025SoK: Machine Unlearning for Large Language Modelscited by 11
38 more
  1. 2025Bradley-Terry and Multi-Objective Reward Modeling Are Complementarycited by 11
  2. 2025Does quantization affect models' performance on long-context tasks?cited by 10
  3. 2026VLA-Forget: Vision-Language-Action Unlearning for Embodied Foundation Modelscited by 10
  4. 2025Improving LLM Unlearning Robustness via Random Perturbationscited by 9
  5. 2025Keeping an Eye on LLM Unlearning: The Hidden Risk and Remedycited by 9
  6. 2026RAZOR: Ratio-Aware Layer Editing for Targeted Unlearning in Vision Transformers and Diffusion Modelscited by 8
  7. 2025A Survey on Collaborating Small and Large Language Models for Performance, Cost-effectiveness, Cloud-edge Privacy, and Trustworthinesscited by 7
  8. 2025Revealing the Deceptiveness of Knowledge Editing: A Mechanistic Analysis of Superficial Editingcited by 5
  9. 2025Exposing Privacy Risks in Graph Retrieval-Augmented Generationcited by 5
  10. 2025Reference-Specific Unlearning Metrics Can Hide the Truth: A Reality Checkcited by 5
  11. 2026Investigating Social Bias Changes in Quantized Language Modelscited by 5
  12. 2026From Signal Degradation to Computation Collapse: Uncovering the Two Failure Modes of LLM Quantizationcited by 5
  13. 2025DUSK: Do Not Unlearn Shared Knowledgecited by 4
  14. 2025Unlearning Inversion Attacks for Graph Neural Networkscited by 4
  15. 2025Unlearning's Blind Spots: Over-Unlearning and Prototypical Relearning Attackcited by 4
  16. 2025SUA: Stealthy Multimodal Large Language Model Unlearning Attackcited by 4
  17. 2025Downgrade to Upgrade: Optimizer Simplification Enhances Robustness in LLM Unlearningcited by 4
  18. 2025A Survey on Unlearning in Large Language Modelscited by 4
  19. 2025Towards Benchmarking Privacy Vulnerabilities in Selective Forgetting with Large Language Modelscited by 4
  20. 2026Hair-Trigger Alignment: Black-Box Evaluation Cannot Guarantee Post-Update Alignmentcited by 4
  21. 2026REBEL: Hidden Knowledge Recovery via Evolutionary-Based Evaluation Loopcited by 4
  22. 2025Align-then-Unlearn: Embedding Alignment for LLM Unlearningcited by 3
  23. 2025Module-Aware Parameter-Efficient Machine Unlearning on Transformerscited by 3
  24. 2025Fine-Tuned LLMs Know They Don't Know: A Parameter-Efficient Approach to Recovering Honestycited by 3
  25. 2026Quantization-Robust LLM Unlearning via Low-Rank Adaptationcited by 3
  26. 2026Position: The Term "Machine Unlearning" Is Overused in LLMscited by 3
  27. 2025GUARD: Guided Unlearning and Retention via Data Attribution for Large Language Modelscited by 2
  28. 2025Dual-Space Smoothness for Robust and Balanced LLM Unlearningcited by 2
  29. 2025LLM Unlearning Under the Microscope: A Full-Stack View on Methods and Metricscited by 2
  30. 2026Per-parameter Task Arithmetic for Unlearning in Large Language Modelscited by 2
  31. 2026KUDA: Knowledge Unlearning by Deviating Representation for Large Language Modelscited by 2
  32. 2026Inference-Time Machine Unlearning via Gated Activation Redirectioncited by 2
  33. 2026Natively Unlearnable Large Language Modelscited by 2
  34. 2026Forgetting Only What Matters: Layer-Selective Unlearning toward Robust LLMscited by 2
  35. 2025MIAT: Maneuver-Intention-Aware Transformer for Spatio-Temporal Trajectory Predictioncited by 1
  36. 2025WARP: Weight Teleportation for Attack-Resilient Unlearning Protocolscited by 1
  37. 2026QUAIL: Quantization Aware Unlearning for Mitigating Misinformation in LLMscited by 1
  38. 2026Copyright Detective: A Forensic System to Evidence LLMs Flickering Copyright Leakage Riskscited by 1

50 most cited of 81 All 81 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026Unlearning in LLMs: Methods, Evaluation, and Open Challenges
  2. 2024A Closer Look at Machine Unlearning for Large Language Models
  3. 2024Towards Effective Evaluations and Comparisons for LLM Unlearning Methods
  4. 2024The Frontier of Data Erasure: Machine Unlearning for Large Language Models
  5. 2025Unlearning That Lasts: Utility-Preserving, Robust, and Almost Irreversible Forgetting in LLMs
  6. 2025LLM Unlearning on Noisy Forget Sets: A Study of Incomplete, Rewritten, and Watermarked Data
  7. 2024To Forget or Not? Towards Practical Knowledge Unlearning for Large Language Models
  8. 2024Towards Robust Evaluation of Unlearning in LLMs via Data Transformations