mlsanityData to 14 Oct 2026

What to Forget in Unlearning? Forget Set Curation for Language Models

arXiv 2608.14855 ↗First posted 14 Aug 2026cs.CLcited by 0

On the map: LLM safety and security → Machine unlearning

Abstract (arXiv)

Machine unlearning aims to remove targeted data or behaviors from a trained model without retraining from scratch. Yet most evaluations assume that the examples to forget are already known. In realistic language-model deployments, a requester may ask a model to stop reproducing a song or book without knowing which spans, documents, quotations, or near-duplicates in a trillion-token corpus support that behavior. We study this missing upstream problem, forget set curation: mapping a suppression request to the data passed to an unlearning algorithm. We introduce CleanSlate, a benchmark for verbatim output suppression over songs and books, with model-specific extraction profiles, content-grounded QA, and capability-retention evaluations. CleanSlate exposes two failure modes. Natural lexical and exact-substring curators often yield forget sets that lead to weak suppression. An evaluation-aware curator suppresses requested continuations almost completely, but causes collateral regression on non-requested content and model-dependent capability loss. These results show that practical unlearning is not only an optimization problem once a forget set is given: the data chosen for forgetting determines both what can be unlearnt and what else is damaged.

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 35 of its 40 references that are in the map. First posted on arXiv 14 Aug 2026.

Builds on

35 of 40 references.

Closest ancestors · 18

Same direction, newest first

  1. 2026Randomized Antipodal Search Done Right for Data Pareto Improvement of LLM Unlearning17 Apr
  2. 2026Extracting books from production language models6 Jan
  3. 2025LLM Unlearning Without an Expert Curated Dataset8 Aug
  4. 2025Distributional Machine Unlearning via Selective Data Removal20 Jul
  5. 2025OpenUnlearning: Accelerating LLM Unlearning via Unified Benchmarking of Methods and Metrics14 Jun
  6. 2025Not All Tokens Are Meant to Be Forgotten3 Jun
  7. 2025Not Every Token Needs Forgetting: Selective Unlearning to Limit Change in Utility in Large Language Model Unlearning1 Jun
  8. 2025LLM Unlearning Reveals a Stronger-Than-Expected Coreset Effect in Current Benchmarks14 Apr
  9. 2024Attribute-to-Delete: Machine Unlearning via Datamodel Matching30 Oct
  10. 2024Evaluating Deep Unlearning in Large Language Models19 Oct
  11. 2024Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning9 Oct
  12. 2024Erasing Conceptual Knowledge from Language Models3 Oct
6 more
  1. 2024MUSE: Machine Unlearning Six-Way Evaluation for Language Models8 Jul
  2. 2024RWKU: Benchmarking Real-World Knowledge Unlearning for Large Language Models16 Jun
  3. 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning8 Apr
  4. 2024UNDIAL: Self-Distillation with Adjusted Logits for Robust Unlearning in Large Language Models15 Feb
  5. 2024TOFU: A Task of Fictitious Unlearning for LLMs11 Jan
  6. 2023Large Language Model Unlearning14 Oct

Other references · 17

  1. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  2. 2020Extracting Training Data from Large Language Modelscited by 2,189
  3. 2022Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Themcited by 1,807
  4. 2017Understanding Black-box Predictions via Influence Functionscited by 1,741
  5. 2023Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generationcited by 1,391
  6. 2022Quantifying Memorization Across Neural Language Modelscited by 763
  7. 2018CoQA: A Conversational Question Answering Challengecited by 683
  8. 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearningcited by 509
  9. 2019Making AI Forget You: Data Deletion in Machine Learningcited by 465
  10. 2022Datamodels: Predicting Predictions from Training Datacited by 189
  11. 2024Infini-gram: Scaling Unbounded n-gram Language Models to a Trillion Tokenscited by 153
  12. 2024DsDm: Model-Aware Dataset Selection with Datamodelscited by 105
5 more
  1. 2024Measuring memorization in language models via probabilistic extractioncited by 41
  2. 2025Extracting memorized pieces of (copyrighted) books from open-weight language modelscited by 30
  3. 2025OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokenscited by 27
  4. 2025Language Models May Verbatim Complete Text They Were Not Explicitly Trained Oncited by 21
  5. 2025Infini-gram mini: Exact n-gram Search at the Internet Scale with FM-Indexcited by 14

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Confess What You Know: Forget-Set Misalignment with Model Knowledge in LLM Unlearning
  2. 2025BLUR: A Benchmark for LLM Unlearning Robust to Forget-Retain Overlap
  3. 2025LLM Unlearning on Noisy Forget Sets: A Study of Incomplete, Rewritten, and Watermarked Data
  4. 2026Learning What to Forget: Improving LLM Unlearning via Learned Token-Level Importance
  5. 2026SHRED: Retain-Set-Free Unlearning via Self-Distillation with Logit Demotion
  6. 2023Unlearn What You Want to Forget: Efficient Unlearning for LLMs
  7. 2025The Erasure Illusion: Stress-Testing the Generalization of LLM Forgetting Evaluation
  8. 2025Unlearning That Lasts: Utility-Preserving, Robust, and Almost Irreversible Forgetting in LLMs