Machine Unlearning

We consider Representation Misdirection (RM), a class of large language model (LLM) unlearning methods that achieve forgetting by redirecting the latent representations of forget-samples toward a target vector. Despite being important, the roles of the target vector used in RM, however, remain underexplored. Here, we approach and revisit RM through the lens of the Linear Representation Hypothesis. Specifically, if one can identify a one-dimensional representation corresponding to a high-level concept, the Linear Representation Hypothesis enables linear operations on this concept vector within the forget-representation space. Under this view, we hypothesize that, beyond forgetting, machine unlearning via RM elicits controllable side effect behaviors and capabilities corresponding to the high-level concept. Our hypothesis is empirically validated across a wide range of concepts and tasks, including controlling unlearned models' truthfulness, sentiment, stereotypical bias, refusal, language, and in-context learning (ICL) tasks. Our findings reveal that this phenomenon could be either a hidden risk if misused or a mechanism that can be harnessed for developing unlearned models that require stronger capabilities and controllable behaviors.

Machine unlearning for large language models (LLMs) remains challenging because full retraining is costly, while approximate methods often struggle to remove targeted behaviors without degrading retained utility, especially under limited post-deployment supervision. We consider a practical PEFT setting for targeted behavioral contamination removal with a small forget set, a limited retain buffer, and LoRA-only updates, and propose RapidUn, an influence-guided framework that converts cross-sample influence estimates into fixed sample-specific weights for weighted LoRA unlearning. Across Llama-3-8B on Dolly-15k and Alpaca-57k, with cross-model validation on Mistral-7B + Dolly-15k, RapidUn achieves lower seen-trigger and OOD-trigger-family ASR than Fisher, GA, and LoReUn while maintaining competitive clean utility. On Llama-3-8B + Alpaca-57k, it achieves a 77x wall-clock speedup over the clean-corpus LoRA retraining reference. Complementary TOFU, semantic LLM-judge, and IFEval evaluations further support the effectiveness of influence-guided sample reweighting beyond the controlled trigger benchmark.

Real-world deployment of text-to-image diffusion models requires continual concept removal as new privacy, copyright, or safety obligations arise over time. Existing unlearning methods, however, are designed for single-step deletion and collapse after only 3-5 sequential applications. We trace this instability to two compounding factors: (i) coarse mapping targets that cause degradation to accumulate unnecessarily across steps, and (ii) the absence of local protection for semantically neighboring concepts, whose shared internal representations make them the first to suffer collateral damage. Because this damage is strongest in the local semantic neighborhood of the forget concept, global replay alone cannot prevent it. Building on this analysis, we propose Locality-Aware Continual Unlearning (LACU), a framework with two complementary mechanisms. Locality-Aware Target Selection chooses, for each forget prompt, the context-preserving mapping prompt that the diffusion model itself treats as most similar to the original prompt, measured by score-prediction distance (how differently the model denoises the same noisy image under two text conditions), ensuring each update is as small and targeted as possible. Locality-Aware Replay uses the same metric to identify the retain concepts closest to the forget concept in the model's own representation and replays them as a local functional regularizer, directly shielding the most vulnerable neighborhood. Combined with teacher-student distillation and lightweight $\ell_2$ parameter regularization, LACU maintains stable unlearning over 10 sequential steps, preserving significantly higher related retention ($RR_{\text{acc}}$) and general retention ($GR_{\text{acc}}$) than recent baselines.

Multimodal large language models (MLLMs) can inadvertently memorize privacy-sensitive information during training. While existing unlearning methods can remove such content, they often severely degrade the model's foundational capabilities, such as general image understanding. This critical shortfall motivates our investigation into benign memory forgetting, the precise removal of targeted, privacy-sensitive knowledge while rigorously preserving unrelated capabilities. To pioneer and evaluate progress toward this objective, we introduce S-MLLMUn Bench, the first benchmark designed to jointly and quantitatively assess an unlearning method's efficacy in knowledge erasure and the preservation of image understanding. Furthermore, we propose the Sculpted Memory Forgetting Adapter (SMFA), a new framework that enables benign memory forgetting. SMFA confines forgetting to designated memory regions, maintaining overall model performance. By initially fine-tuning the model to replace sensitive outputs with refusals, SMFA generates a memory forgetting adapter, followed by a retaining anchor-guided masking mechanism that safeguards unrelated knowledge. Extensive experiments on S-MLLMUn Bench demonstrate that existing methods fail to achieve benign forgetting, whereas our proposed SMFA serves as an effective baseline, successfully achieving targeted knowledge erasure without compromising the model's foundational visual capabilities. Code and data are available at https://github.com/zeng-zhen/S-MLLMUn.

Graph unlearning (GU) has emerged as a promising solution to comply with "the right to be forgotten" regulations by enabling the removal of sensitive information upon request. However, this solution is not foolproof. The involvement of multiple parties creates new attack surfaces, and residual traces of deleted data can persist within the unlearned graph neural networks (GNNs). These vulnerabilities can be exploited by attackers to recover the supposedly erased samples, undermining the intended functionality of GU. In this work, we propose GraphToxin, the first full graph reconstruction attack against GU. We show that GraphToxin can recover not only a deleted individual's information and personal links but also sensitive content of their neighbors, thereby posing substantially more detrimental threats than prior membership inference attacks (MIAs). Specifically, we introduce a novel curvature matching module to provide fine-grained guidance for full unlearned graph recovery. We further extend GraphToxin to multiple-node removal under both white-box and black-box settings, showcasing its practical feasibility and potential to cause considerable harm. We highlight the necessity of worst-case analysis and propose a systematic evaluation framework to assess attack performance under both random and worst-case node removal scenarios. Our extensive experiments demonstrate the effectiveness and flexibility of GraphToxin. Notably, current defense mechanisms are largely ineffective against this attack. Additionally, our findings reveal that existing GU verification standards based on MIAs can be misleading: they achieve membership-level protection, while the full unlearned graph remains recoverable through GraphToxin.

Data unlearning aims to remove the influence of specific training samples from a trained model. In fine-tuning methods, data unlearning relies primarily on loss maximization over forget samples, which often leads to quality degradation or incomplete forgetting. Existing methods perform unlearning uniformly across diffusion stages, ignoring diffusion dynamics from noise to data. Our systematic study of diffusion phases shows that forgetting in diffusion models is uneven across time and frequency, with theoretical justification of distributive distortion and forgetting-utility trade-off. By selectively forgetting time and frequency in diffusion models, we achieve both higher unlearning success rates and improved generation quality across diverse settings, including both conditional and unconditional scenarios. We also introduce an improved SSCD metric that measures dissimilarity using a normalized perturbation distance. Together, we provide practical insights for understanding and improving data unlearning in diffusion models.

Speech Emotion Recognition (SER) infers a speaker's emotional state from speech and is increasingly deployed in human-computer interaction, education, and healthcare. Because speech also carries sensitive personal information, speakers may ask that some of their recordings be deleted, which requires removing the influence of those samples from an already trained SER model. Most machine unlearning methods can meet this request only with access to the remaining training data alongside the samples to be forgotten; this is impractical when the remaining data cannot be redistributed or has itself been deleted, and it adds storage and computation as the data grows. To this end, we propose a retain-free unlearning method that updates a pre-trained SER model using only the forget set. Our key idea is to synthesise adversarial samples from the forget set as a surrogate for the unavailable remaining data, and to constrain each parameter update by its estimated importance so that forgetting does not erase general knowledge. The experiments over several emotional-speech corpora and self-supervised backbones show that our method drives forget-set performance down to near chance while retaining much of the model's utility on the remaining and unseen-speaker data, narrowing the gap to methods that rely on the remaining set.

Large language models deployed in sensitive applications increasingly require the ability to unlearn specific knowledge, such as user requests, copyrighted materials, or outdated information, without retraining from scratch to ensure regulatory compliance, user privacy, and safety. This task, known as machine unlearning, aims to remove the influence of targeted data (forgetting) while maintaining performance on the remaining data (retention). A common approach is to formulate this as a multi-objective problem and reduce it to a single-objective problem via scalarization, where forgetting and retention losses are combined using a weighted sum. However, this often results in unstable training dynamics and degraded model utility due to conflicting gradient directions. To address these challenges, we propose OFMU, a penalty-based bi-level optimization framework that explicitly prioritizes forgetting while preserving retention through a hierarchical structure. Our method enforces forgetting via an inner maximization step that incorporates a similarity-aware penalty to decorrelate the gradients of the forget and retention objectives, and restores utility through an outer minimization step. To ensure scalability, we develop a two-loop algorithm with provable convergence guarantees under both convex and non-convex regimes. We further provide a rigorous theoretical analysis of convergence rates and show that our approach achieves better trade-offs between forgetting efficacy and model utility compared to prior methods. Extensive experiments across vision and language benchmarks demonstrate that OFMU consistently outperforms existing unlearning methods in both forgetting efficacy and retained utility.

Large language models (LLMs) are trained on massive datasets that may include private or copyrighted content. Due to growing privacy and ownership concerns, data owners may request the removal of their data from trained models. Machine unlearning provides a practical solution by removing the influence of specific data without full retraining. However, most existing methods still suffer from over-unlearning due to the lack of a principled mechanism to regulate the forgetting boundary, leading to unnecessary utility degradation and heightened privacy and robustness risks. In this work, we propose EGUP (Entanglement-Guided Unlearning with Proxy Constraint), a novel framework that leverages entanglement and proxy constraint to guide the unlearning process while mitigating over-unlearning. Within each iteration, EGUP employs inter-sample entanglement to adaptively reweight the unlearning strength, assigning greater unlearning efforts to forget samples that are semantically closer to retained knowledge. Across iterations, EGUP leverages intra-sample entanglement to track the representation shift of each forget sample and dynamically adjust its unlearning effort. In addition, we incorporate a proxy constraint that approximates the model's expected outputs after unlearning, forming a reference boundary that softly regularizes the unlearning process. EGUP is compatible with existing gradient-based objectives and serves as a plug-and-play enhancement. We evaluate EGUP on the TOFU and MUSE benchmarks, demonstrating consistent improvements in the unlearning-utility trade-off across multiple LLMs. Moreover, EGUP achieves performance close to the retrained model while remaining scalable and robust.

Privacy protection laws, such as the GDPR, grant individuals the right to request the forgetting of their personal data not only from databases but also from machine learning (ML) models trained on them. Machine unlearning has emerged as a practical means to facilitate model forgetting of data instances seen during training. Although some existing machine unlearning methods guarantee exact forgetting, they are typically costly in computational terms. On the other hand, more affordable methods do not offer forgetting guarantees and are applicable only to specific ML models. In this paper, we present efficient unlearning with privacy guarantees (EUPG), a novel machine unlearning framework that offers formal privacy guarantees to individuals whose data are being unlearned. EUPG involves pre-training ML models on data protected using privacy models, and it enables {\em efficient unlearning with the privacy guarantees offered by the privacy models in use}. Through empirical evaluation on four heterogeneous data sets protected with $k$-anonymity and $ε$-differential privacy as privacy models, our approach demonstrates utility and forgetting effectiveness comparable to those of exact unlearning methods, while significantly reducing computational and storage costs. Our code is available at https://github.com/najeebjebreel/EUPG.

Unlearning large language models (LLMs) is important due to copyright and privacy concerns. A key challenge is unintended forgetting, where removing data inadvertently impairs model utility. While prior work has focused on architectural innovations, the influence of data-level factors remains underexplored. We introduce GUARD, a framework for Guided Unlearning And Retention via Data Attribution. GUARD uses a lightweight proxy data attribution metric tailored to LLM unlearning, which efficiently quantifies the "alignment" between the Forget and Retain (or synthetic supplement) Sets. This metric leads to a new unlearning objective that assigns adaptive, nonuniform weights to samples, decaying exponentially with their proxy attribution scores, and thereby provably mitigates unintended retention loss. On the TOFU benchmark with a 10% forget set, applying GUARD to Gradient Ascent (GA) improves Retain Set ROUGE-L from 0.18 to 22.40 (a 124x increase) and Retain Probability from 0.06% to 17.50%, a substantial recovery from the GA collapse. Furthermore, GUARD combined with NPO (Negative Preference Optimization) outperforms Representation Misdirection for Unlearning (RMU) with 33.80% and 53.10% reductions in the Retain Set Sacrifice Rates computed from ROUGE-L and Probability ($ρ_r$ and $ρ_p$), and a 16.01% improvement in knowledge retention on the MUSE NEWS Retain Set. These findings persist when attribution scores are computed using synthetic or highly subsampled (20%) Retain Sets, where NPO+GUARD still reduces $ρ_r$ and $ρ_p$ by up to 43.89% and 48.13% over NPO. Similar improvements are observed on MUSE, with a tolerable absolute privacy leakage degradation of 13.7 compared to the best-performing existing method in the privacy category.

As generative models become increasingly powerful and pervasive, the ability to unlearn specific data due to privacy concerns, legal requirements, or harmful content has become increasingly important. Unlearning aims to selectively remove the influence of particular data without costly retraining from scratch. Effective unlearning requires (i) forgetting undesired data, (ii) preserving generation quality, (iii) retaining the influence of desired training data on model parameters, and (iv) computational efficiency. We propose fast unlearning algorithms based on loss gradient orthogonalization for unconditional and conditional generative models. We consider the loss function of the original model restricted to the retain data and augment it by a term promoting the orthogonality between the two gradients of the loss function evaluated at the retain dataset and the forget dataset. We coin this method unlearning via orthogonalization (UNO) and show that the additional weak orthogonality constraint accelerates unlearning. The inclusion of the weak orthogonality constraint implies that each gradient descent step utilizes local information about the geometry via the Hessian of the loss function. Our geometry aware gradient descent step significantly accelerates the learning of the retain data loss function compared to a gradient descent which uses only gradient information. By alternating between an UNO and a gradient surgery step, we obtain UNO-S, which further reduces execution time. We demonstrate that our algorithms effectively forget data while preserving the fidelity of the original model. On standard image benchmarks, they achieve orders-of-magnitude faster unlearning than existing methods such as gradient surgery. We validate our approach on datasets of increasing complexity (MNIST, CelebA, and ImageNet-1K) and generative architectures ranging from VAEs to diffusion transformers.

Despite significant progress in safety alignment, large language models (LLMs) remain susceptible to jailbreak attacks. Existing defense mechanisms have not fully deleted harmful knowledge in LLMs, which allows such attacks to bypass safeguards and produce harmful outputs. To address this challenge, we propose a novel safety alignment strategy, Constrained Knowledge Unlearning (CKU), which focuses on two primary objectives: knowledge localization and retention, and unlearning harmful knowledge. CKU works by scoring neurons in specific multilayer perceptron (MLP) layers to identify a subset U of neurons associated with useful knowledge. During the unlearning process, CKU prunes the gradients of neurons in U to preserve valuable knowledge while effectively mitigating harmful content. Experimental results demonstrate that CKU significantly enhances model safety without compromising overall performance, offering a superior balance between safety and utility compared to existing methods. Additionally, our analysis of neuron knowledge sensitivity across various MLP layers provides valuable insights into the mechanics of safety alignment and model knowledge editing.

Diffusion models excel at generating high-quality images but can memorize and reproduce harmful concepts when prompted. Although fine-tuning methods have been proposed to unlearn a target concept, they struggle to fully erase it while maintaining generation quality on other concepts, leaving models vulnerable to jailbreak attacks. Existing jailbreak methods demonstrate this vulnerability but offer limited insight into how unlearned models retain harmful concepts, limiting progress on effective defenses. In this work, we show that a coherent, interpretable, attack-accessible linear residual of the erased concept can be recovered in the token embedding space, and that both an attack and a defense follow from this structure. We introduce SubAttack, a novel jailbreaking attack that reads out this subspace by learning an orthogonal set of attack token embeddings, each being a linear combination of human-interpretable textual elements, revealing that unlearned models still retain the target concept through related textual components. Furthermore, our attack is also more powerful and transferable across text prompts, initial noises, and unlearned models than prior attacks. Conversely, projecting out the same subspace yields SubDefense, a lightweight plug-and-play defense mechanism that suppresses the residual concept in unlearned models. SubDefense provides stronger robustness than existing defenses while better preserving safe generation quality. Extensive experiments across multiple unlearning methods, concepts, and attack types demonstrate that our approach advances both understanding and mitigation of vulnerabilities in diffusion unlearning.

Machine unlearning is a promising approach to improve LLM safety by removing unwanted knowledge from the model. However, prevailing gradient-based unlearning methods suffer from issues such as high computational costs, hyperparameter instability, poor sequential unlearning capability, vulnerability to relearning attacks, low data efficiency, and lack of interpretability. While Sparse Autoencoders are well-suited to improve these aspects by enabling targeted activation-based unlearning, prior approaches underperform gradient-based methods. This work demonstrates that, contrary to these earlier findings, SAEs can significantly improve unlearning when employed dynamically. We introduce $\textbf{Dynamic DAE Guardrails}$ (DSG), a novel method for precision unlearning that leverages principled feature selection and a dynamic classifier. Our experiments show DSG substantially outperforms leading unlearning methods, achieving superior forget-utility trade-offs. DSG addresses key drawbacks of gradient-based approaches for unlearning -- offering enhanced computational efficiency and stability, robust performance in sequential unlearning, stronger resistance to relearning attacks, better data efficiency including zero-shot settings, and more interpretable unlearning.

Approximate unlearning has gained popularity as an approach to efficiently update an LLM so that it behaves (roughly) as if it was not trained on a subset of data to begin with. However, existing methods are brittle in practice and can easily be attacked to reveal supposedly unlearned information. To alleviate issues with approximate unlearning, we instead propose SIFT-Masks (SIgn-Fixed Tuning-Masks), an exact unlearning method based on model merging. SIFT-Masks addresses two key limitations of standard model merging: (1) merging a large number of tasks can severely harm utility; and (2) methods that boost utility by sharing extra information across tasks make exact unlearning prohibitively expensive. SIFT-Masks solves these issues by (1) applying local masks to recover task-specific performance; and (2) constraining finetuning to align with a global sign vector as a lightweight approach to determine masks independently before merging. Across four settings where we merge up to 500 models, SIFT-Masks improves accuracy by 5-80% over naive merging and uses up to 250x less compute for exact unlearning compared to other merging baselines.

How can we effectively remove or ``unlearn'' undesirable information, such as specific features or the influence of individual data points, from a learning outcome while minimizing utility loss and ensuring rigorous guarantees? We introduce a unified mathematical framework based on information-theoretic regularization to address both data-point unlearning and feature unlearning. For data-point unlearning, we introduce the Marginal Unlearning Principle, an auditable and provable framework. Moreover, we provide an information-theoretic unlearning definition based on the proposed principle and provable guarantees on sufficiency and necessity of marginal unlearning. We then show that the proposed framework provides a natural solution to the marginal unlearning problem and yields auditable high-probability marginal-unlearning guarantees. For feature unlearning, the framework applies to deep learning with flexible training objectives. By combining flexibility in learning objectives with simplicity in regularization design, our approach is highly adaptable and practical for a wide range of machine learning and AI applications. From a mathematical perspective, we provide a unified analytic solution to the optimal feature unlearning problem with a variety of information-theoretic training objectives. Our theoretical analysis reveals intriguing connections between machine unlearning, information theory, optimal transport, and extremal sigma algebras. Numerical simulations support our theoretical findings.

Machine unlearning, the process of selectively removing data from trained models, is increasingly crucial for addressing privacy concerns and knowledge gaps post-deployment. Despite this importance, existing approaches are often heuristic and lack formal guarantees. In this paper, we analyze the fundamental utility, time, and space complexity trade-offs of approximate unlearning, providing rigorous certification analogous to differential privacy. For in-distribution forget data -- data similar to the retain set -- we show that a surprisingly simple and general procedure, empirical risk minimization with output perturbation, achieves tight unlearning-utility-complexity trade-offs, addressing a previous theoretical gap on the separation from unlearning "for free" via differential privacy, which inherently facilitates the removal of such data. However, such techniques fail with out-of-distribution forget data -- data significantly different from the retain set -- where unlearning time complexity can exceed that of retraining, even for a single sample. To address this, we propose a new robust and noisy gradient descent variant that provably amortizes unlearning time complexity without compromising utility.

Current research on deep machine unlearning primarily focuses on improving or evaluating the overall effectiveness of unlearning methods while overlooking the varying difficulty of unlearning individual training samples. As a result, the broader feasibility of machine unlearning remains under-explored. This paper studies the cruxes that make machine unlearning difficult through a thorough instance-level unlearning performance analysis over various unlearning algorithms and datasets. In particular, we summarize four factors that make unlearning a data point difficult, and we empirically show that these factors are independent of a specific unlearning algorithm but only relevant to the target model and its training data. Given these findings, we argue that machine unlearning research should pay attention to the instance-level difficulty of unlearning.

Machine unlearning aims to remove the influence of specified training data or knowledge from a trained model without requiring full retraining. This capability is particularly important for modern image classifiers and large language models (LLMs), where retraining can be computationally expensive. However, machine unlearning on complex data remains difficult because the target information is often distributed across multiple semantic elements. Existing methods mainly remove samples, modify labels, or edit model parameters to reduce the influence of the forgetting target. These approaches usually do not explicitly identify which semantic concepts connect the forgetting target to the model's prediction or generated response. As a result, it is difficult to determine which information to modify. This uncertainty may leave residual target information or unnecessarily affect knowledge that should be retained. To address this gap, we propose a concept-guided poisoning unlearning framework that explicitly identifies the concepts that contribute most strongly to the target class or knowledge and uses them to guide the model update. For image classification, our method first identifies class-relevant concepts with a Post-hoc Concept Bottleneck Model, localizes the image regions that express these concepts, and constructs replacement-based poisoned samples. For LLMs, it elicits the target knowledge through multiple questions, aggregates Integrated Gradients across the resulting responses to identify consistently important content, and masks this content to construct poisoned training targets. Experiments on multiple datasets show that the proposed framework achieves effective unlearning across different tasks while largely preserving retained model utility.

Jailbreaking attacks can enable Large Language Models (LLMs) to bypass the safeguard and generate harmful content. Existing jailbreaking defense methods have failed to address the fundamental issue that harmful knowledge resides within the model, leading to potential jailbreak risks for LLMs. In this paper, we propose a novel defense method called Eraser, which mainly includes three goals: unlearning harmful knowledge, retaining general knowledge, and maintaining safety alignment. The intuition is that if an LLM forgets the specific knowledge required to answer a harmful question, it will no longer have the ability to answer harmful questions. The training of Erase does not actually require the model's own harmful knowledge, and it can benefit from unlearning general answers related to harmful queries, which means it does not need assistance from the red team. The experimental results show that Eraser can significantly reduce the jailbreaking success rate for various attacks without compromising the general capabilities of the model. Our codes are available at https://github.com/ZeroNLP/Eraser.