Machine Unlearning Evaluation

Machine unlearning in large language models aims to remove unwanted knowledge while preserving the model's remaining capabilities. Although existing methods use retention objectives or restrict where edits occur, achieving the desired forgetting level can still leave collateral changes that impair non-target behavior. Our recovery comparisons suggest that some of these changes can be reversed while preserving observed forgetting performance. In this work, we present Propose-Then-Project Unlearning (PTP-U), a framework that combines targeted forgetting with the recovery of non-target capabilities. PTP-U first applies local analytic edits to weaken target knowledge associations, then aligns non-target output distributions with those of the original model to recover capabilities while maintaining fixed forgetting constraints. Both stages serve a common goal: satisfying the forgetting requirements while preserving fluent generation and performance on non-target tasks. Across three benchmarks, PTP-U achieves the strongest forgetting-retention trade-off among evaluated methods, reaching 81.22%-91.03% forgetting while preserving 94.20% non-target utility on average. At matched forgetting, PTP-U consistently retains higher non-target utility.

As multimodal large language models (MLLMs) become more capable and widely deployed, concerns about privacy and safety have become increasingly pressing. Machine unlearning offers one approach to addressing these concerns by removing designated information from trained models while preserving unrelated capabilities. However, fragmented implementations and evaluation protocols, incomplete robustness testing, and limited understanding of metric reliability make progress in MLLM unlearning difficult to assess systematically. We introduce Open-MMUnlearning, an open-source, extensible framework that integrates target-model preparation, multimodal data processing, unlearning, and evaluation through shared interfaces and structured configurations. The framework supports five benchmarks spanning privacy, safety, and copyright, eight MLLMs from four model families, and twelve unlearning methods. Its evaluation suite jointly assesses forgetting effectiveness, retained utility, and robustness to model interventions, adversarial inputs, and membership inference attacks. Using a common evaluation protocol, we compare ten representative unlearning methods. In this comparison, GD and MIP-Editor tie for the highest overall score: GD achieves the highest Forget Quality, while MIP-Editor preserves more Model Utility. We further introduce a metric meta-evaluation protocol that tests faithfulness using models with controlled exposure to target knowledge and robustness under quantization and relearning. Among the thirteen evaluated metrics, BLEU achieves the highest aggregate reliability score. KS-Test attains the highest faithfulness AUC but performs less well on robustness. Together, the framework and these findings support reproducible comparison of MLLM unlearning methods and systematic assessment of evaluation reliability.

Exact unlearning requires a deployed predictor to match one rebuilt without the information named by a deletion request. Existing general-purpose exact methods localize retraining through disjoint shards, but every request still invalidates a model, and smaller shards reduce the data available to each constituent predictor. We introduce Quantized Sufficient Statistics (QSS), which separates a small frozen schema from mutable, sum-decomposable content. The schema learns global structure; the content stores local prediction corrections as additive statistics indexed by quantized regions. Deleting content is therefore exact subtraction rather than optimization. We distinguish two guarantees: QSS-L exactly removes a label while retaining the unlabelled input, whereas QSS-E exactly removes both input and label by learning the schema without deletable examples. A deletion takes the arithmetic fast path with probability $1-ρ$ and triggers a full rebuild with probability $ρ$; all reported expected latencies include both events. Across 15 vision, text, and tabular datasets at $ρ=0.5\%$, QSS-L is within 2 percentage points of SISA on 11 tasks and provides 4--483$\times$ lower expected deletion latency on the low-class-count tasks where a compact schema is effective. QSS-E quantifies the additional accuracy cost of removing every trace of an input.

Existing attacks on unlearned diffusion models assume that the erased concepts are known in advance and focus on recovering them. In practice, however, model providers may not disclose which concepts have been removed, and even with access to the original base model, an adversary may still lack a clear target to attack. In this paper, we aim to answer the following critical but overlooked questions: which concepts have been erased from the model, and how many have been erased in total? To this end, we present Tracer, a framework that rapidly and accurately identifies erased concepts and estimates their number. Tracer efficiently identifies erased concepts without generating and classifying images. By combining lightweight spectral analysis of weight footprints, it enables efficient search over large candidate vocabularies. To distinguish multiple erased concepts, we introduce a footprint coverage objective that guides sequential discovery. Tracer estimates the number of erased concepts by detecting a sharp decline in candidate confidence as the selected concepts account for the erasure footprint, without requiring labeled examples for calibration. The framework requires only lightweight linear algebra and limited forward probes, with no prior knowledge of the unlearning algorithm. Experiments across text-to-image and text-to-video backbones and diverse unlearning methods demonstrate that Tracer identifies erased concepts and estimates their number in seconds, achieving 150 to 137,000 times and 133 to 20,000 times speedups over MIA and brute-force search on image and video models, respectively, with substantially higher identification accuracy.

Unlearning a fact in one language does not guarantee its removal in others as changing the query or even the requested answer language can reopen seemingly forgotten knowledge -- a cross-lingual loophole. The most straightforward solution to this challenge -- unlearning in all languages -- is neither scalable nor desirable as it amplifies damage to unrelated model capabilities. We introduce the task of language budgeted multilingual unlearning where the goal is to select a subset of languages that maximizes cross-lingual erasure. To study this task we introduce the Cross-Lingual Unlearning Tensor, an unlearning benchmark that spans 174 language--script pairs and 25 atomic paraphrase types to examine when forgetting generalizes across linguistic expressions of the same knowledge. We further propose COVER, which selects source languages to maximize predicted COVERage of languages receiving no forget supervision, enabling unlearning on a language budget. Surprisingly, we find naively selecting strong individual sources does not reliably compose into strong source sets motivating our development of COVER. At deployment COVER only requires benign calibration data and access to the frozen model. Across three model families and two disjoint forget sets, COVER reduces mean held-out residual access by 7.8--27.3% relative to uniform source selection. We find these gains extend beyond synthetic benchmarks to real news documents in low-resource language settings using human translated data from the Low Resource Languages for Emergent Incidents (LORELEI) corpus.

In materials machine learning, closely related retained structures can sustain accurate property predictions even after removing a specific record, rendering post-deletion prediction error an ambiguous metric for machine unlearning. To resolve this ambiguity, we define the deletion floor as the expected target loss under a specified retraining procedure at the deleted request. Standard indistinguishability constraints yield a sharp interval bounding an update's target loss around this baseline reference. Theoretically, a conditional neighbor bound links a low deletion floor directly to retained fit, prediction regularity, and local label agreement, while an exact ridge identity isolates residual fit from the prediction change induced by record deletion. Empirically, controlled redundancy sweeps show an $\approx 8\times$ drop in median normalized retraining loss when one retained relative remains after deletion. Across two distinct fitting regimes in a paired Materials Project study, the lower-floor regime also exhibits a larger prediction change on more than 50% of the shared requests. Systematic comparisons against approximate updates and the original model decouple deliberate target suppression from preserved overall model utility. Consequently, request-level unlearning evaluations should report reference loss, prediction change, and retained utility together, interpreting post-deletion accuracy against what retraining itself leaves behind.

The rising number of concept unlearning techniques for text-to-image (T2I) diffusion models has produced a fragmented evaluation landscape. Methods are assessed under heterogeneous experimental conditions making principled cross-method comparison difficult. We present eval-unlearn, an open-source Python library providing a unified, reproducible benchmarking framework for concept unlearning in T2I Diffusion models. eval-unlearn integrates twelve published unlearning techniques spanning fine-tuning, closed-form model editing, and inference-time intervention, alongside nine complementary evaluation metrics covering erasure efficacy, adversarial robustness, generative quality, and concept retention. Its plugin architecture lets third-party techniques and metrics self-register without modifying the core framework, and its streaming, batched pipeline supports efficient evaluation of both standard NSFW concepts and arbitrary general concepts. As a further contribution, we release a public leaderboard on HuggingFace along with an interactive tool for real-time evaluation of unlearning techniques. The leaderboard compares nudity concept erasure case study across all twelve techniques, exposing significant accuracy-quality trade-offs that are obscured by heterogeneous evaluation. eval-unlearn is released under the MIT license; the package, code, leaderboard, and documentation are all available at https://eval-unlearn.readthedocs.io.

Machine unlearning (MU) offers a path to compliance with "right to be forgotten" regulations. While MU has received increasing attention for speech tasks, it remains largely unexplored for Automatic Speech Recognition (ASR). In this work, we investigate whether existing MU algorithms and evaluation tools are suitable for ASR. We apply several MU techniques to an ASR model, evaluating privacy-utility trade-offs for single-subject unlearning, then assess the best algorithm under sequential and simultaneous unlearning. Results show that gradient ascent-based algorithms achieve strong utility-privacy trade-offs, whereas more complex approaches over-unlearn samples, making them easier to identify as unlearned. This suggests standard privacy evaluations based on simple Membership Inference attacks are insufficient to reliably assess unlearning success, motivating improved evaluation methods for MU in ASR. Finally, we show that both sequential and simultaneous unlearning yield worse privacy and utility than single-subject unlearning, underscoring the need for unlearning constructions better suited to these settings.

Diffusion data-point unlearning is typically evaluated immediately after each deletion, even though subsequent requests may repeatedly update the same model. We identify sequential reappearance, a failure mode in which an instance that is initially judged to be forgotten later returns to the memorized regime without reuse of the deleted data or adversarial fine-tuning. To capture this behavior, we introduce a target-level evaluation protocol that tracks whether each target is forgotten immediately, remains forgotten at the end of the sequence, or reappears during subsequent deletions. We further find that targets that later reappear exhibit sharper local denoising-loss geometry after deletion than targets that remain forgotten.

We evaluate Deep Perturbation Learning (DPL), which perturbs training images and labels along influence-derived directions, in three roles in which prior work has positioned it for machine unlearning: a direct deletion signal (the strongest claim), a utility-preserving regularizer, and a warm start for adversarial unlearning. Evidence for the weaker roles has been used to support the stronger one, so we test each role separately under a matched protocol with exact-seed retraining baselines. An audit of the public implementation identifies two correctness issues: image directions are computed on augmented, normalized tensors but applied to raw images, and the label perturbation falls below float32 resolution, leaving labels unchanged. After correcting the image-perturbation pipeline, DPL fails the direct-deletion criterion on CIFAR-10/ResNet-18 in all three paired seeds. Its utility effects are inconsistent in sign across seeds, and once direction-computation time is counted it underperforms simple warm-start baselines. A one-seed Tiny ImageNet check likewise does not favor DPL as a regularizer or warm start; preprocessing inconsistencies in the released code make the direct comparison there inconclusive. These results cover random instance deletion only and do not rule out influence-based methods in other deletion regimes. We release a role-matched evaluation protocol and an audit checklist for perturbation-based deletion claims.

Unlearning benchmarks such as TOFU and MUSE certify forgetting by reading the model's final answer, where a model that refuses to answer already counts as having forgotten. We show that this model-level certificate does not transfer once the model is deployed as an agent. We introduce K-Bench, a benchmark that scores LLM unlearning under agentic deployment. K-Bench inspects all six channels a ReAct agent exposes, including its chain-of-thought (CoT), tool calls and tool observations, and elicited summary. A query counts as leaked if the secret appears in any of them. Each experiment places the secret in exactly one of the agent's three sources (the weights, the prompt, or the retrieval store). The K-Score is computed separately for each source and credits forgetting only when the agent remains usable. Clearing the answer channel does not make the secret unrecoverable. On structured retrieval, the secret stays verbatim in the tool-observation channel and the aggregate leak rate is unchanged. When the secret lives in the prompt or the retrieval store, TOFU and MUSE report no leakage, while the deployed agent still leaks it on 22--86% of queries. When the secret is in the weights, none of the twenty evaluated published methods demonstrably removes it, and only an input-corruption intervention reaches selective forgetting under the evaluated observer. The top-ranked method changes across base models. A refusal-tuning method resists the evaluated extraction without verified knowledge removal.

Noisy labels remain a critical challenge for training deep neural networks, since memorizing incorrect labels degrades generalization. Once noisy samples are identified after training, the standard solution is to retrain the model from scratch on the cleaned dataset, which is increasingly expensive as datasets and models grow. Machine Unlearning (MU) has recently emerged as a computationally efficient alternative, but the relative effectiveness of different MU strategies for noisy-label correction remains poorly understood. In this work, we conduct a comparative empirical study of five MU methods (NegGrad, Fine-Tuning (FT), Random Labeling (RL), SalUn, and MUNBa) across symmetric, asymmetric, instance-dependent, and open-set noise on CIFAR-10, CIFAR-100, and the real-world noisy dataset Food-101N. Our central finding is that the appropriate unlearning strategy is conditioned on the noise structure. Simple FT is a strong baseline across most closed-set scenarios; RL and SalUn are the most consistently robust methods and, under instance-dependent noise, approach retraining accuracy at a fraction of the computational cost; MUNBa shows advantages mainly under extreme symmetric noise. Under open-set noise, in contrast, we show that retraining on the cleaned subset degrades accuracy relative to the noisy baseline, so approximating the retrained model is not an adequate objective in this regime. On Food-101N, all MU methods remain competitive and achieve accuracies close to retraining despite reducing runtime by an order of magnitude. These findings provide practical guidelines for selecting MU strategies for post-training noisy-label correction.

Recently, machine unlearning, the removal of specific training data influence from a model, has gained increasing attention. In large language models (LLMs), unlearning is particularly challenging due to the ambiguity of inputs and outputs. Con- sequently, rigorous evaluation is critical for assessing both safety and utility, and for driving progress in unlearning meth- ods. We identify two key shortcomings in existing unlearning benchmarks: (1) they do not actively test whether unlearned information can still be forcibly extracted, and (2) they fail to evaluate performance preservation on boundary questions, be- nign queries that are semantically close to the unlearned con- tent. Here we introduce WMDP++, an extension of WMDP that addresses these gaps by incorporating targeted extrac- tion of unlearned information and systematic evaluation on boundary questions. WMDP++ provides a more stringent and informative benchmark for evaluating unlearning in LLMs.

Prior white-box studies show that large language models can retain latent traces of target knowledge after unlearning, even when the knowledge is no longer expressed in their outputs. However, existing audits remain limited to one-off diagnostics: it is unclear whether these residual signals can predict future recovery under continued training or serve as reliable optimization targets. Resolving this gap is essential to determine whether internal auditing can move beyond post-hoc evaluation toward proactive risk monitoring and safer unlearning. We propose J-Access, an inference-time audit that uses the Jacobian lens to map intermediate representations into vocabulary space and measures how often target concepts remain accessible along the model's output pathway. We hypothesize that residual accessibility reflects recovery susceptibility: knowledge that remains closer to the output pathway requires less fine-tuning to restore, leading to faster recovery. We audit 398 public unlearned models spanning eight unlearning methods. We find that: (1) most unlearned models retain access above the retain-only gold level; (2) pre-attack accessibility predicts recovery speed and extent at the model level, but cannot identify which specific facts will be recovered; and (3) directly minimizing J-Access does not promote genuine deletion. Instead, the model learns to hide knowledge from the audit, producing lower audit scores but greater post-attack recovery. These findings position J-Access as a model-level diagnostic for assessing residual susceptibility in unlearned models. We argue internal audits should serve as an independent diagnostic dimension in unlearning evaluation, and should not be converted into optimization targets without validation.

Multimodal large language models (MLLMs) have demonstrated remarkable capabilities in storing and recalling rich person-related knowledge, raising increasing concerns about reliable knowledge removal. However, existing machine unlearning approaches for MLLMs typically assume access to original forget and retain corpora, which are often unavailable in realistic deletion scenarios. To address this limitation, we introduce PRMU, a benchmark for evaluating corpus-free multimodal unlearning under realistic person-centric deletion requests. PRMU focuses on naturally acquired person-related knowledge and evaluates whether models can remove target knowledge while preserving related knowledge through diverse textual and visual probes, including adversarial evaluation and fine-grained locality analysis. To facilitate research in this setting, we further introduce Similarity-Gated Projection Editing (SGPE), a lightweight corpus-free unlearning baseline with knowledge displacement, protected parameter-space editing, and locality-aware multimodal control. Extensive experiments on representative MLLMs reveal that existing unlearning methods often suffer from unfavorable forgetting-locality trade-offs, with significant locality degradation under aggressive forgetting settings, and remain vulnerable to multimodal knowledge reactivation. Meanwhile, SGPE provides a competitive trade-off between target forgetting, locality preservation, and general multimodal utility. We hope PRMU can facilitate future research toward realistic and scalable multimodal machine unlearning. Code and dataset will be released at https://github.com/2231122/PRMU.

Benchmarking machine unlearning methods is critical to understand whether sensitive knowledge is removed from large language models (LLMs) or not. Current unlearning benchmarks include mainly single-hop questions and a narrow set of multi-hop questions. Although effective, they still face two challenges. (1) Knowledge is not isolated, whereby diverse multi-hop reasoning paths can potentially induce knowledge leakage than normal queries. (2) Unlearning may be fragile: unlearned knowledge can be partially recovered through recovery attacks such as lightweight post-unlearning adaptation, making static evaluation insufficient. Therefore, in this paper, we introduce \unlearning as a novel benchmark to understand robust LLM knowledge removal across diverse reasoning paths and recovery attacks. We experiment with this benchmark on 3 models, 6 unlearning methods, and 2 carefully curated datasets. Results show that existing methods are vulnerable to multi-hop reasoning paths and recovery attacks. We further explore the trade-off among forget quality, robustness, and model utility for LLM unlearning.

Vision-Language Models (VLMs), like Large Language Models (LLMs), may memorize sensitive, copyrighted, or harmful knowledge from their pretraining corpora. Removing such knowledge is essential for building trustworthy AI systems. However, existing studies primarily focus on forgetting within individual modalities. Although recent work has begun to explore cross-modal consistency in unlearning, the cross-modal transfer of real-world knowledge unlearning remains insufficiently studied. To address this gap, we introduce UNLINK-VL, a real-world benchmark for cross-modal knowledge unlearning in VLMs. Under a post-hoc unlearning setting in which the original forget and retain corpora are unavailable, UNLINK-VL selects visually identifiable real-world entities as unlearning targets and associates them with corresponding images and one-hop and multi-hop facts derived from Wikidata. The benchmark comprises four complementary subsets that evaluate direct forgetting of target knowledge, the propagation of forgetting through relational knowledge, the preservation of related non-target knowledge, and robustness to semantically equivalent queries. We train models under text-only and multimodal unlearning settings and evaluate forgetting effectiveness and retained utility across textual, visual, and cross-modal scenarios. Extensive experiments reveal a pronounced asymmetry in cross-modal transfer: multimodal unlearning remains effective under textual evaluation, whereas text-only unlearning transfers poorly to visual and cross-modal scenarios. Meanwhile, the evaluated methods largely preserve the models' general capabilities. These findings demonstrate that relying solely on intra-modal evaluation, particularly text-only evaluation, may substantially overestimate the effectiveness of knowledge unlearning in VLMs, underscoring the need for cross-modal unlearning and evaluation.

Machine unlearning offers a promising approach to remove unsafe content from Multimodal Large Language Models (MLLMs), yet ensuring the precision of unlearning remains a persistent challenge. One reason is that current MLLM unlearning evaluation paradigms suffer from a critical blind spot: they assess model utility through benchmarks whose representations are distant from the forget set, failing to capture knowledge holes---severe degradation on benign adjacent inputs. To probe knowledge holes in unlearned MLLMs, we construct a benchmark that captures unintended degradation on benign inputs sharing generic patterns with the forget set, and confirm through controlled experiments that they are a systematic consequence of commonly used approaches. Furthermore, to bridge this gap, we propose Selective Protection with Anchored Regularization, which protects generic patterns via anchored activation filtering while reinforcing them through entity-abstracted enhancement. Our experiments on SafeEraser demonstrate that SPAR recovers over 98% of vanilla response quality compared to below 50% for standard baselines---while achieving 0.00% attack success rate and competitive model utility. These results underscore the necessity of more fine-grained evaluation for trustworthy MLLM unlearning.

Machine unlearning removes the influence of user-specified training examples from a trained model, avoiding the need to retrain it from scratch. Localization-based methods improve unlearning efficiency by identifying a subset of influential model parameters. However, existing approaches select parameters based solely on forget-set importance, neglecting their role in retained dataset and often causing collateral damage to semantically similar retained examples. We address this limitation with a retain-aware localization method that considers parameter importance to both forgotten and retained data. We also introduce a retain-similar evaluation set, constructed using cosine similarity in the model embedding space, to directly measure collateral damage. Across eleven experimental settings on CIFAR-10 dataset and ResNet18 model, our method consistently reduces collateral damage while improving standard unlearning metrics, demonstrating the effectiveness of retain-aware localization for similarity-aware machine unlearning.

Large language model unlearning is consistently fragile under relearn attacks. On TOFU, fine-tuning on twenty forget examples substantially recovers held-out forget-set ROUGE for every method we evaluate, and we trace this fragility to optimization geometry. The per-token answer margin of fourteen post-hoc methods spanning gradient, preference, and distillation families converges into a narrow band above the retain reference in 41 of 42 method--size cells, a regularity we call the margin cliff. We prove that this cliff follows whenever the retain coupling holds the diagnostic log-odds of forget content above a floor, a condition that token-saturating losses induce at stationarity and that we verify directly on 34 of 42 cells. Margin Calibration (\textsc{MC}) is a plug-in polish adding a non-saturating margin hinge anchored at the reference's per-token margin plus a KL probe on a disjoint instruction corpus, restoring forget-side pressure where the native loss saturates. Under a stated gradient-dominance condition, whose on-trajectory gradient signature we measure by instrumenting the polish, its stationary set lies on the cliff-crossing side, yielding an attack-budget upper bound on the relearn margin lift. Across TOFU (three Llama-3 sizes, three forget tiers), MUSE-News on Llama-2-7B-hf, and a Phi-3.5 panel, a single frozen configuration wins all 14 head-to-head forget aggregates and all populated relearn cells (panel-mean post-attack ROUGE-L $0.41$ to $0.18$) and lowers raw membership AUC on 13/14, with reduced retain-side utility as the main cost. A deployment variant matches these gains without a retain-trained reference.

Concept unlearning is increasingly used to limit the reproduction of protected or unsafe visual concepts in text-to-image models. Existing evaluations, however, mostly study targets that dominate the whole image, such as styles, broad object categories, or portrait-like identities, leaving company logos comparatively underexamined. Logos create a different failure mode: a small localized mark can carry the entire protected concept, must be visually precise to remain recognizable, and can be triggered implicitly by products, storefronts, packaging, or advertisements even when the word ``logo'' is absent. We introduce LU-500, a logo-unlearning benchmark built from Fortune Global 500 companies to study this localized and semantically entangled setting. LU-500 contains nearly 10,000 curated text-query and logo-image pairs, with an explicit track (LUex-500) and an implicit contextual track (LUim-500). To avoid reducing the task to a binary detector score, we define a multi-grained protocol that evaluates both local logo removal and global image preservation in pixel and latent spaces. Experiments on representative inference-time methods, including NP, SLD, and SEGA, and compatible fine-tuning-based methods such as ESD and Forget-Me-Not, show that the evaluated methods struggle to remove logo evidence without changing non-target content. We further analyze ProLU, a prompt-space multi-agent baseline: it improves local erasure by removing logo-inducing semantics, but also illustrates why prompt filtering is not a substitute for weight-level disentanglement. Correlation analyses over logo area, location, and structural complexity suggest that future logo unlearning may need spatially aware controls, such as SSIM-guided constraints, rather than purely global concept suppression.

Machine unlearning has emerged as a tool for removing personal data from trained models to comply with recent AI regulations. To evaluate unlearning effectiveness in multimodal large language models (MLLMs), prior works fine-tune models on fictitious identities, simulating unlearning requests on subsets of these IDs, which are typically uniformly distributed. However, in realistic scenarios, people from different demographic groups may request to be unlearned at different frequencies, potentially altering the model's internal beliefs for these groups and leading to biased behaviors. To fill this gap, we propose FAIRGET, the first Visual Question Answering benchmark that evaluates unlearning under unbalanced, realistic, forget requests. These requests are designed to simulate multiple realistic scenarios, ranging from simple to challenging settings, that lead to biased unlearned models if fairness is not accounted for. Additionally, we propose FAUN, the first unlearning algorithm for MLLMs that forgets unlearning data while preserving model fairness. FAUN exploits a bias-aware activation steering mechanism to unlearn identities while accounting for the unbalanced nature of the forget data. Experiments on FAIRGET and the established FIUBench demonstrate our method's superiority both in unlearning quality and fairness.

Machine unlearning is commonly evaluated by matching a retrained oracle on trained probes. In a controlled nonce-fact testbed with a matched retraining reference, we find this criterion can favor methods that retain held-out knowledge: candidates it rates adequate score held-out forget facts $-2.82$ nats below the never-learned level (cluster CI $[-3.16,-2.48]$). We recast unlearning as restoration to the matched reference and audit oracle-free screens and certificate-style criteria across 45 model-seed cells spanning five open architecture families. The reference itself falsifies an absolute retain/round-trip certificate: the injected model, which retains the retain set by construction, fails the fixed retain threshold in 41/45 cells and its own round trip in 31/45, and the reference fully certifies in only 1/45. A base-anchored held-out screen remains strong as a selective necessary test: on a sealed challenge suite it rejects the injected model in 45/45 cells, accepts the reference in 44/45, and partially detects entity-routing suppression (35/45); it is a necessary test with measured sensitivity, not a sufficiency certificate. A damage-relative recalibration anchored to the reference's own operating point certifies a small subset in 15/45 cells; where it does not abstain, its picks lie within retraining noise (0.80 nats) on the axes it optimizes, while the common trained-probe criterion sits 5.17 nats away (a supporting comparison, not a head-to-head benchmark). A fixed-magnitude logit-suppression attack defeats the full forward battery in 12/45 cells, so forward-only certification is not sound; our method is an empirical selective test for methods-as-produced. An identifiability theorem delimits which facts admit an oracle-free forget threshold at all, with TOFU as the predicted boundary case.

Machine unlearning in LLMs is the targeted removal of specific knowledge while preserving all other capabilities, critical for privacy and safety. Yet existing benchmarks measure it unreliably. They miss knowledge that resurfaces under paraphrased or indirect queries, a failure we call under-forgetting, and lack the semantic, syntactic, and lexical probes needed to verify that unrelated knowledge is preserved, a failure we call over-forgetting. Both failures reflect an asymmetric generalization problem. Forget evaluation must cover diverse query formulations of the same target facts, testing whether forgetting holds beyond exact training prompts. Retain evaluation must probe a far larger and implicitly defined set, namely every fact disjoint from the forget target. The retain set thus defines the effective forget set, yet current datasets provide no fine-grained annotation of this forget-retain boundary. We address this with SUITE, an evaluation protocol and training corpus that captures forget-retain structure for real-world factual domains. Methods trained on SUITE improve substantially, showing that training data is as important as algorithmic design. Building on the obtained insights, we introduce JensUn++, an unlearning algorithm that achieves the best forget-retain utility trade-off across three LLMs, in both sequential and joint unlearning settings. Code and datasets are available at https://amitpeleg.github.io/forget-narrowly-retain-broadly

Evaluating whether unlearning algorithms truly remove training data influence remains an open challenge. We propose a practical auditor that computes data-dependent lower bounds on the unlearning parameter $\varepsilon$ using membership inference attacks. Evaluating multiple unlearning algorithms, we find a sharp separation: algorithms with rigorous guarantees, such as model clipping and rewind-to-delete, achieve very small $\varepsilon$ bounds that do not falsify their unlearning guarantees, whereas empirical methods such as Hessian-based unlearning, interleaved ascent-descent, ascent on the forget set, and fine-tuning on the retain set exhibit large bounds, indicating poor unlearning. Our auditor provides a practical tool for empirically falsifying unlearning claims through a hypothesis-testing framework, and we validate it on CIFAR-100 and Shakespeare text.

Association unlearning aims to disable learned label-attribute shortcuts while preserving task performance. Existing evaluations mainly measure output-level robustness or probe whether shortcut attributes remain readable in frozen features, but neither test determines whether a retained association remains functionally usable by the original classifier. We propose the Association Restoration Test (ART), a post-hoc diagnostic for functional shortcut restorability. ART estimates class-conditional association directions, amplifies residual components, and evaluates the modified features with the original classifier head. Across Waterbirds, CelebA, SpuCoDogs, and an ISIC timestamp-artifact extension, we show that output metrics, representation probes, and ART characterize distinct aspects of shortcut mitigation. These findings motivate restoration-aware evaluation for unlearning and shortcut-mitigation methods that target learned associations rather than individual classes or concepts.

Multimodal Large Language Models (MLLMs) have shown strong capabilities, but they may memorize private information from web data, raising privacy concerns. Machine unlearning offers a way to remove such private knowledge without retraining from scratch. However, existing MLLM unlearning benchmarks have two major limitations. First, they rely on simplified images that contain only the single target individual, failing to reflect the visual complexity of real-world photos. Second, they typically assume that the forget set and retain set are fully separated, ignoring the fact that private information is often visually entangled with benign public information. For example, a private individual may appear with a public figure or in front of a well-known landmark, where unlearning the private target should not damage the public context. To address these limitations, we propose PPE-Bench, a new benchmark for evaluating MLLM unlearning under private-public entanglement. Each image contains a target individual to be forgotten and public information to be preserved, including public figure and landmark. We further introduce two simple but effective methods to better preserve public information during unlearning. Through experiments, we find that existing unlearning methods can reduce private information leakage, but often substantially harm adjacent public information.

LLMs memorize sensitive training data, including personally identifiable information (PII), creating a pressing need for reliable post hoc removal methods. Unlearning has emerged as a promising solution, with state-of-the-art(SOTA) methods often following a localize-first, unlearn-second paradigm that targets specific model parameters. However, existing benchmarks evaluate unlearning solely at the output level, leaving open the question of whether unlearning truly erases knowledge from a model's parameters or merely obfuscates it, a concern reinforced by the success of resurfacing attacks. To bridge this gap, we introduce LACUNA: the first unlearning testbed with ground-truth parameter-level localization. LACUNA injects PII of synthetic individuals into predefined parameters of 1B and 7B OLMo-based models via masked continual pretraining, enabling direct evaluation of whether unlearning targets the weights responsible for knowledge storage. We use LACUNA to benchmark current SOTA unlearning methods and find that, despite strong output-level performance, existing methods are highly imprecise and susceptible to resurfacing attacks. We further show that when localization is successful, even a simple gradient-based unlearning method achieves strong erasure and robustness to resurfacing attacks, highlighting the importance of precise unlearning. We release LACUNA to complement behavioral evaluations and drive further advances in robust, localization-based unlearning.

Machine unlearning (MU) is commonly judged by output forgetting, such as low forget-set accuracy or reduced logit-level membership inference. But if output-level success can coexist with retraining-inconsistent residuals in representation space, what kind of forgetting are current evaluations actually certifying? We study this question through retraining-consistent representation forgetting, using the retrained model (i.e., trained from scratch without the forget data) as an operational reference for correct forgetting. Across multiple unlearning methods, datasets, and models, our theoretical analysis and empirical results show that standard output-level evaluation can systematically overestimate the success of unlearning. Under this stronger lens, current methods often appear forgotten at the output layer while exhibiting a structured mismatch relative to retraining. They partially align with retraining on forget samples, remain more inconsistent on retain samples, and leave residual discrepancy concentrated along retraining-related directions rather than diffuse in representation space. This structured mismatch is characterized by forget/retain asymmetry, directional mismatch, and concentrated residuals along retraining-related directions. These results suggest that current MU is often evaluated for apparent forgetting rather than retraining-consistent forgetting. More broadly, retraining reveals what output forgetting hides.

Machine unlearning for large language models (LLMs) aims to remove specified knowledge while preserving the rest of the model's capabilities. However, the boundary between knowledge to forget and knowledge to retain is often unclear, since related and even distant information may be entangled in the model. In this paper, we study LLM unlearning from a data-centric perspective and measure how unlearning effects propagate from the forget set to same-domain and distant-domain knowledge not after but before unlearning. We find a consistent decay pattern: collateral damage is strongest near the forget set, weakens with semantic distance, but does not disappear at domain boundaries. We further ask whether such damage can be audited before unlearning is executed. We formulate forget-set auditing as a pre-unlearning prediction task and analyze which data features are most predictive of downstream damage. Our results show that interaction features between the forget set and evaluation set provide the strongest signals, suggesting that collateral damage is partly reflected in data geometry before model updates occur. These findings position forget-set auditing as an early warning tool for identifying risky unlearning runs and designing more reliable unlearning procedures. Code and data are available at https://github.com/BartSu/PreUnlearn.