Machine Unlearning Evaluation

Large language model (LLM)-based agents have recently gained considerable attention due to the powerful reasoning capabilities of LLMs. Existing research predominantly focuses on enhancing the task performance of these agents in diverse scenarios. However, as LLM-based agents become increasingly integrated into real-world applications, significant concerns emerge regarding their accumulation of sensitive or outdated knowledge. Addressing these concerns requires the development of mechanisms that allow agents to selectively forget previously learned knowledge, giving rise to a new term LLM-based agent unlearning. This paper initiates research on unlearning in LLM-based agents. Specifically, we propose a novel and comprehensive framework that categorizes unlearning scenarios into three contexts: state unlearning (forgetting specific states or items), trajectory unlearning (forgetting sequences of actions) and environment unlearning (forgetting entire environments or categories of tasks). Within this framework, we introduce a natural language-based unlearning method that trains a conversion model to transform high-level unlearning requests into actionable unlearning prompts, guiding agents through a controlled forgetting process. Moreover, to evaluate the robustness of the proposed framework, we introduce an unlearning inference adversary capable of crafting prompts, querying agents, and observing their behaviors in an attempt to infer the forgotten knowledge. Experimental results show that our approach effectively enables agents to forget targeted knowledge while preserving performance on untargeted tasks, and prevents the adversary from inferring the forgotten knowledge.

As multimodal models like CLIP become integral to downstream systems, the need to remove sensitive information is critical. However, machine unlearning for contrastively-trained encoders remains underexplored, and existing evaluations fail to diagnose fine-grained, association-level forgetting. We introduce SALMUBench (Sensitive Association-Level Multimodal Unlearning), a benchmark built upon a synthetic dataset of 60K persona-attribute associations and two foundational models: a Compromised model polluted with this data, and a Clean model without it. To isolate unlearning effects, both are trained from scratch on the same 400M-pair retain base, with the Compromised model additionally trained on the sensitive set. We propose a novel evaluation protocol with structured holdout sets (holdout identity, holdout association) to precisely measure unlearning efficacy and collateral damage. Our benchmark reveals that while utility-efficient deletion is feasible, current methods exhibit distinct failure modes: they either fail to forget effectively or over-generalize by erasing more than intended. SALMUBench sets a new standard for comprehensive unlearning evaluation, and we publicly release our dataset, models, evaluation scripts, and leaderboards to foster future research.

Federated Unlearning (FUL) aims to remove specific participants' data contributions from a trained Federated Learning model, thereby ensuring data privacy and compliance with regulatory requirements. Despite its potential, progress in FUL has been limited due to several challenges, including the cross-client knowledge inaccessibility and high computational and communication costs. To overcome these challenges, we propose Federated On-server Unlearning (FOUL), a novel framework that comprises two key stages. The learning-to-unlearn stage serves as a preparatory learning phase, during which the model identifies and encodes the key features associated with the forget clients. This stage is communication-efficient and establishes the basis for the subsequent unlearning process. Subsequently, on-server knowledge aggregation phase aims to perform the unlearning process at the server without requiring access to client data, thereby preserving both efficiency and privacy. We introduce a new data setting for FUL, which enables a more transparent and rigorous evaluation of unlearning. To highlight the effectiveness of our approach, we propose a novel evaluation metric termed time-to-forget, which measures how quickly the model achieves optimal unlearning performance. Extensive experiments conducted on three datasets under various unlearning scenarios demonstrate that FOUL outperforms the Retraining in FUL. Moreover, FOUL achieves competitive or superior results with significantly reduced time-to-forget, while maintaining low communication and computation costs.

Unlearning in Large Language Models (LLMs) aims to enhance safety, mitigate biases, and comply with legal mandates, such as the right to be forgotten. However, existing unlearning methods are brittle: minor query modifications, such as multi-hop reasoning and entity aliasing, can recover supposedly forgotten information. As a result, current evaluation metrics often create an illusion of effectiveness, failing to detect these vulnerabilities due to reliance on static, unstructured benchmarks. We propose a dynamic framework that stress tests unlearning robustness using complex structured queries. Our approach first elicits knowledge from the target model (pre-unlearning) and constructs targeted probes, ranging from simple queries to multi-hop chains, allowing precise control over query difficulty. Our experiments show that the framework (1) shows comparable coverage to existing benchmarks by automatically generating semantically equivalent Q&A probes, (2) aligns with prior evaluations, and (3) uncovers new unlearning failures missed by other benchmarks, particularly in multi-hop settings. Furthermore, activation analyses show that single-hop queries typically follow dominant computation pathways, which are more likely to be disrupted by unlearning methods. In contrast, multi-hop queries tend to use alternative pathways that often remain intact, explaining the brittleness of unlearning techniques in multi-hop settings. Our framework enables practical and scalable evaluation of unlearning methods without the need for manual construction of forget test sets, enabling easier adoption for real-world applications. We release the pip package and the code at https://sites.google.com/view/unlearningmirage/home.

Machine unlearning aims to remove the influence of specific training data from pre-trained models without retraining from scratch, and is increasingly important for large language models (LLMs) due to safety, privacy, and legal concerns. Although prior work primarily evaluates unlearning in static, single-turn settings, forgetting robustness under realistic interactive use remains underexplored. In this paper, we study whether unlearning remains stable in interactive environments by examining two common interaction patterns: self-correction and dialogue-conditioned querying. We find that knowledge appearing forgotten in static evaluation can often be recovered through interaction. Although stronger unlearning improves apparent robustness, it often results in behavioral rigidity rather than genuine knowledge erasure. Our findings suggest that static evaluation may overestimate real-world effectiveness and highlight the need for ensuring stable forgetting under interactive settings.

Evaluating machine unlearning remains challenging, as existing methods typically require retraining reference models or performing membership inference attacks, both of which rely on prior access to training configuration or supervision labels, making them impractical in realistic scenarios. Motivated by the fact that most unlearning algorithms remove a small, random subset of the training data, we propose a subset-level evaluation framework based on statistical independence. Specifically, we design a tailored use of the Hilbert-Schmidt Independence Criterion to assess whether the model outputs on a given subset exhibit statistical dependence, without requiring model retraining or auxiliary classifiers. Our method provides a simple, standalone evaluation procedure that aligns with unlearning workflows. Extensive experiments demonstrate that our approach reliably distinguishes in-training from out-of-training subsets and clearly differentiates unlearning effectiveness, even when existing evaluations fall short.

The need for machine unlearning is critical for data privacy, yet existing methods often cause Knowledge Contamination by unintentionally damaging related knowledge. Such a degraded model performance after unlearning has been recently leveraged for new inference and backdoor attacks. Most studies design adversarial unlearning requests that require poisoning or duplicating training data. In this study, we introduce a new unlearning-induced attack model, namely indirect unlearning attack, which does not require data manipulation but exploits the consequence of knowledge contamination to perturb the model accuracy on security-critical predictions. To mitigate this attack, we introduce a theoretical framework that models neural networks as Neural Knowledge Systems. Based on this, we propose ROKA, a robust unlearning strategy centered on Neural Healing. Unlike conventional unlearning methods that only destroy information, ROKA constructively rebalances the model by nullifying the influence of forgotten data while strengthening its conceptual neighbors. To the best of our knowledge, our work is the first to provide a theoretical guarantee for knowledge preservation during unlearning. Evaluations on various large models, including vision transformers, multi-modal models, and large language models, show that ROKA effectively unlearns targets while preserving, or even enhancing, the accuracy of retained data, thereby mitigating the indirect unlearning attacks.

Machine unlearning (MU) refers to the post-training capability to remove (the influence of) training examples that are incorrect, biased, or leak sensitive/private information. MU is now widely regarded as critical for building safe and fair AI. In parallel, research into transformer architectures for computer vision has been highly successful: Vision Transformers (VTs) increasingly emerge as strong alternatives to CNNs. Yet, MU research for vision tasks has largely centered on CNNs, not VTs. While MU benchmarks have been developed for LLMs, diffusion models, and CNNs, none currently exist for VTs. This work is the first to attempt this, benchmarking MU algorithm performance across different VT families (ViT, Swin-T, and DINOv2) and at different capacities. The work employs (i) different datasets, selected to assess the impacts of dataset scale and complexity; (ii) different MU algorithms, selected to represent fundamentally different approaches for MU; and (iii) both single-shot and continual unlearning protocols. Additionally, it focuses on benchmarking MU algorithms that leverage training data memorization, since leveraging memorization has been recently discovered to significantly improve the performance of previously SOTA algorithms. En route, the work characterizes how VTs memorize training data relative to CNNs, and assesses the impact of different memorization proxies on performance. The benchmark uses unified evaluation metrics that capture two complementary notions of forget quality along with accuracy on unseen (test) data and on retained data. Overall, this work offers a benchmarking basis, enabling reproducible, fair, and comprehensive comparisons of existing (and future) MU algorithms on VTs. Importantly, for the first time, it sheds light on how well existing algorithms work in VT settings, establishing a promising reference performance baseline.

Machine Unlearning (MU) enables Large Language Models (LLMs) to remove unsafe or outdated information. However, existing work assumes that all facts are equally forgettable and largely ignores whether the forgotten knowledge originates from pretraining or supervised fine-tuning (SFT). In this paper, we introduce DUET (Dual Unlearning Evaluation across Training Stages), a benchmark of 28.6k Wikidata-derived triplets annotated with fact popularity using Wikipedia link counts and LLM-based salience scores. Our experiments show that pretrained and SFT models respond differently to unlearning. An SFT step on the forget data yields smoother forgetting, more stable tuning, and 10-50% higher retention, while direct unlearning on pretrained models remains unstable and prone to relearning or catastrophic forgetting.

While many recent methods aim to unlearn or remove knowledge from pretrained models, seemingly erased knowledge often persists and can be recovered in various ways. Because large foundation models are far from interpretable, understanding whether and how such knowledge persists remains a significant challenge. To address this, we turn to the recently developed framework of combinatorial interpretability. This framework, designed for two-layer neural networks, enables direct inspection of the knowledge encoded in the model weights. We reproduce baseline unlearning methods within the combinatorial interpretability setting and examine their behavior along two dimensions: (i) whether they truly remove knowledge of a target concept (the concept we wish to remove) or merely inhibit its expression while retaining the underlying information, and (ii) how easily the supposedly erased knowledge can be recovered through various fine-tuning operations. Our results shed light within a fully interpretable setting on how knowledge can persist despite unlearning and when it might resurface.

Evaluating machine unlearning methods remains technically challenging, with recent benchmarks requiring complex setups and significant engineering overhead. We introduce a unified and extensible benchmarking suite that simplifies the evaluation of unlearning algorithms using the KLoM (KL divergence of Margins) metric. Our framework provides precomputed model ensembles, oracle outputs, and streamlined infrastructure for running evaluations out of the box. By standardizing setup and metrics, it enables reproducible, scalable, and fair comparison across unlearning methods. We aim for this benchmark to serve as a practical foundation for accelerating research and promoting best practices in machine unlearning. Our code and data are publicly available.

As pretrained models are increasingly shared on the web, ensuring that models can forget or delete sensitive, copyrighted, or private information upon request has become crucial. Machine unlearning has been proposed to address this challenge. However, current evaluations for unlearning methods rely on output-based metrics, which cannot verify whether information is completely deleted or merely suppressed at the representation level, where suppression is insufficient for true unlearning. To address this gap, we propose a novel restoration-based analysis framework that uses Sparse Autoencoders to identify class-specific expert features in intermediate layers and applies inference-time steering to quantitatively distinguish between suppression and deletion. Applying our framework to 12 major unlearning methods in image classification tasks, we find that most methods achieve high restoration rates of unlearned information, indicating that they only suppress information at the decision-boundary level, while preserving semantic features in intermediate representations. Notably, even retraining from pretrained checkpoints shows high restoration, revealing that robust semantic features inherited from pretraining are not removed by retraining. These results demonstrate that representation-level retention poses significant risks overlooked by output-based metrics, highlighting the need for new unlearning evaluation criteria. We propose new evaluation guidelines that prioritize representation-level verification, especially for privacy-critical applications in the era of pre-trained models.

Targeted interventions on language models, such as unlearning or model editing, aim to modify specific information, but their effects often propagate to related, unintended areas (e.g., removing virology content may degrade performance on allergies); these side-effects are commonly referred to as the ripple effect. We introduce RippleBench-Maker, an automatic pipeline that retrieves semantic neighbors of any source concept from a knowledge repository and generates multiple-choice questions at varying semantic distances. We instantiate this framework using WikiRAG, an open-source RAG system over English Wikipedia, to construct RippleBench-WMDP-Bio (584 seed topics, 352,961 questions), and evaluate eight unlearning methods on Llama3-8B-Instruct. All eight exhibit accuracy drops that are largest near the unlearned target and decay with semantic distance, each with a distinct propagation profile. We replicate these findings across Mistral-7B, Zephyr-7B, and Yi-34B; cross-model delta curves are nearly identical, suggesting ripple effects are a property of the unlearning method rather than the base model. We validate all major pipeline stages using a four-experiment Mechanical Turk study (5,200+ responses, 61 workers). We release all code, data, and infrastructure.

Machine learning models are increasingly deployed as software components that must evolve as requirements change. When specific training records or features must no longer influence a deployed model, machine unlearning aims to remove that influence without retraining from scratch. Because unlearning is often approximate, its effectiveness must be tested. Such tests must often treat the model as a black box, without access to its parameters, training history, or unlearning procedure. Features pose a further challenge: even after a feature is removed from a model's inputs, its influence can persist through downstream features. Many existing checks examine only the feature's direct use and can therefore certify a model that still depends on it. We frame unlearning testing as specification-based testing and present CAFÉ, which, using only a deployed model's predictions, intervenes on the feature, propagates the change to its downstream features, and checks whether the predictions still respond. CAFÉ measures a target's residual influence through both its direct and indirect causal paths, and its fine-grained diagnostics show which channels and subgroups still carry it. On two causal-network benchmarks with four unlearning methods, CAFÉ ranks residual influence with 0.92--0.93 pairwise accuracy, against at most 0.71 for existing checks, which fail in both directions: they certify models whose influence persists through downstream features and flag correctly unlearned ones. On real census data, CAFÉ likewise exposes influence that survives retraining yet goes unnoticed by direct-input checks.

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.