# Leak@k: Unlearning Does Not Make LLMs Forget Under Probabilistic Decoding

arXiv: https://arxiv.org/abs/2511.04934 · HTML: https://mlsanity.com/paper/2511.04934/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- Placed in Machine unlearning, one of 12 directions in LLM safety and security.
- Builds on 43 of its 49 references that are in the map.
- First posted on arXiv 7 Nov 2025.

## Placed in

- LLM safety and security / [Machine unlearning](https://mlsanity.com/safety/machine-unlearning/)

## Builds on (most cited first, 30 of 43)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://mlsanity.com/paper/1908.10084/) (2019)
- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [BERTScore: Evaluating Text Generation with BERT](https://mlsanity.com/paper/1904.09675/) (2019)
- [GPT-4o System Card](https://mlsanity.com/paper/2410.21276/) (2024)
- [The Curious Case of Neural Text Degeneration](https://mlsanity.com/paper/1904.09751/) (2019)
- [HuggingFace's Transformers: State-of-the-art Natural Language Processing](https://mlsanity.com/paper/1910.03771/) (2019)
- [A Survey on LLM-as-a-Judge](https://mlsanity.com/paper/2411.15594/) (2024)
- [Hierarchical Neural Story Generation](https://mlsanity.com/paper/1805.04833/) (2018)
- [SemEval-2017 Task 1: Semantic Textual Similarity - Multilingual and Cross-lingual Focused Evaluation](https://mlsanity.com/paper/1708.00055/) (2017)
- [The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning](https://mlsanity.com/paper/2403.03218/) (2024)
- [Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning](https://mlsanity.com/paper/2404.05868/) (2024)
- [Scalable Extraction of Training Data from (Production) Language Models](https://mlsanity.com/paper/2311.17035/) (2023)
- [Who's Harry Potter? Approximate Unlearning in LLMs](https://mlsanity.com/paper/2310.02238/) (2023)
- [Large Language Model Unlearning](https://mlsanity.com/paper/2310.10683/) (2023)
- [TrustLLM: Trustworthiness in Large Language Models](https://mlsanity.com/paper/2401.05561/) (2024)
- [Unrolling SGD: Understanding Factors Influencing Machine Unlearning](https://mlsanity.com/paper/2109.13398/) (2021)
- [MUSE: Machine Unlearning Six-Way Evaluation for Language Models](https://mlsanity.com/paper/2407.06460/) (2024)
- [Gender bias and stereotypes in Large Language Models](https://mlsanity.com/paper/2308.14921/) (2023)
- [Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning](https://mlsanity.com/paper/2410.07163/) (2024)
- [Turning Up the Heat: Min-p Sampling for Creative and Coherent LLM Outputs](https://mlsanity.com/paper/2407.01082/) (2024)
- [Continual Learning and Private Unlearning](https://mlsanity.com/paper/2203.12817/) (2022)
- [Increasing Diversity While Maintaining Accuracy: Text Data Generation with Large Language Models and Human Interventions](https://mlsanity.com/paper/2306.04140/) (2023)
- [Copyright Violations and Large Language Models](https://mlsanity.com/paper/2310.13771/) (2023)
- [Identifying and Mitigating the Security Risks of Generative AI](https://mlsanity.com/paper/2308.14840/) (2023)
- [OpenUnlearning: Accelerating LLM Unlearning via Unified Benchmarking of Methods and Metrics](https://mlsanity.com/paper/2506.12618/) (2025)
- [Towards LLM Unlearning Resilient to Relearning Attacks: A Sharpness-Aware Minimization Perspective and Beyond](https://mlsanity.com/paper/2502.05374/) (2025)
- [LatestEval: Addressing Data Contamination in Language Model Evaluation through Dynamic and Time-Sensitive Test Construction](https://mlsanity.com/paper/2312.12343/) (2023)

## Built on by (most cited first, 9 of 9)

- [REBEL: Hidden Knowledge Recovery via Evolutionary-Based Evaluation Loop](https://mlsanity.com/paper/2602.06248/) (2026)
- [Can an AI Assistant Really Forget? Auditable Deletion from Addressable Memory](https://mlsanity.com/paper/2607.27539/) (2026)
- [Distinguishable Deletion: Unifying Knowledge Erasure and Refusal for Large Language Model Unlearning](https://mlsanity.com/paper/2605.16776/) (2026)
- [Divergence Decoding: Inference-Time Unlearning via Auxiliary Models](https://mlsanity.com/paper/2605.31293/) (2026)
- [LLM Unlearning for Cyber Defense: A Survey on Methods, Challenges, and Emerging Threats](https://mlsanity.com/paper/2607.16227/) (2026)
- [An Empirical Study of Reward Specification and Benchmark Reliability in GRPO-based LLM Unlearning](https://mlsanity.com/paper/2608.17804/) (2026)
- [Graph-Guided Selective Unlearning for Language Models: Controlling Support Routes Beyond Forget Seeds](https://mlsanity.com/paper/2608.26743/) (2026)
- [Do LLMs Really Forget? Hidden-State Leakage in Model Unlearning and How to Fix it](https://mlsanity.com/paper/2609.36612/) (2026)
- [UnlearningSoup: Is Repeated Tuning Necessary for Large Language Model Unlearning?](https://mlsanity.com/paper/2609.37076/) (2026)

## Abstract (arXiv)

Unlearning in large language models (LLMs) is critical for regulatory compliance and for building ethical generative AI systems that avoid producing private, toxic, illegal, or copyrighted content. Despite rapid progress, in this work, we show that textitalmost all existing unlearning methods fail to achieve true forgetting in practice. Specifically, while evaluations of these `unlearned' models under deterministic (greedy) decoding often suggest successful knowledge removal using standard benchmarks, we show that sensitive information reliably resurfaces when models are sampled with standard probabilistic decoding. To rigorously capture this vulnerability, we introduce textttleak@k, a new meta-evaluation metric that quantifies the likelihood of forgotten knowledge reappearing when generating k samples from the model under realistic decoding strategies. Using three widely adopted benchmarks, TOFU, MUSE, and WMDP, we conduct the first large-scale, systematic study of unlearning reliability using textttleak@k metric. Our findings demonstrate that knowledge leakage persists across methods and tasks, underscoring that current state-of-the-art (SOTA) unlearning techniques provide only limited forgetting. We propose an algorithm, termed Robust Unlearning under LEak@k metric (textttRULE) to address this concern. We demonstrate that textttRULE provides an unlearned model for TOFU benchmark with no information leakage for a large number of generation samples. On the MUSE benchmark, textttRULE outperforms SOTA unlearning methods under the textttleak@k metric across most sampling budgets k. Codes are available at https://github.com/OptimAI-Lab/Leak-k.