Rectifying Privacy and Efficacy Measurements in Machine Unlearning: A New Inference Attack Perspective

Machine unlearning focuses on efficiently removing specific data from trained models, addressing privacy and compliance concerns with reasonable costs. Although exact unlearning ensures complete data removal equivalent to retraining, it is impractical for large-scale models, leading to growing interest in inexact unlearning methods. However, the lack of formal guarantees in these methods necessitates the need for robust evaluation frameworks to assess their privacy and effectiveness. In this work, we first identify several key pitfalls of the existing unlearning evaluation frameworks, e.g., focusing on average-case evaluation or targeting random samples for evaluation, incomplete comparisons with the retraining baseline. Then, we propose RULI (Rectified Unlearning Evaluation Framework via Likelihood Inference), a novel framework to address critical gaps in the evaluation of inexact unlearning methods. RULI introduces a dual-objective attack to measure both unlearning efficacy and privacy risks at a per-sample granularity. Our findings reveal significant vulnerabilities in state-of-the-art unlearning methods, where RULI achieves higher attack success rates, exposing privacy risks underestimated by existing methods. Built on a game-based foundation and validated through empirical evaluations on both image and text data (spanning tasks from classification to generation), RULI provides a rigorous, scalable, and fine-grained methodology for evaluating unlearning techniques.

Enhancing One-run Privacy Auditing with Quantile Regression-Based Membership Inference

Differential privacy (DP) auditing aims to provide empirical lower bounds on the privacy guarantees of DP mechanisms like DP-SGD. While some existing techniques require many training runs that are prohibitively costly, recent work introduces one-run auditing approaches that effectively audit DP-SGD in white-box settings while still being computationally efficient. However, in the more practical black-box setting where gradients cannot be manipulated during training and only the last model iterate is observed, prior work shows that there is still a large gap between the empirical lower bounds and theoretical upper bounds. Consequently, in this work, we study how incorporating approaches for stronger membership inference attacks (MIA) can improve one-run auditing in the black-box setting. Evaluating on image classification models trained on CIFAR-10 with DP-SGD, we demonstrate that our proposed approach, which utilizes quantile regression for MIA, achieves tighter bounds while crucially maintaining the computational efficiency of one-run methods.

Winter Soldier: Backdooring Language Models at Pre-Training with Indirect Data Poisoning

The pre-training of large language models (LLMs) relies on massive text datasets sourced from diverse and difficult-to-curate origins. Although membership inference attacks and hidden canaries have been explored to trace data usage, such methods rely on memorization of training data, which LM providers try to limit. In this work, we demonstrate that indirect data poisoning (where the targeted behavior is absent from training data) is not only feasible but also allow to effectively protect a dataset and trace its use. Using gradient-based optimization prompt-tuning, we make a model learn arbitrary secret sequences: secret responses to secret prompts that are absent from the training corpus. We validate our approach on language models pre-trained from scratch and show that less than 0.005% of poisoned tokens are sufficient to covertly make a LM learn a secret and detect it with extremely high confidence ($p < 10^{-55}$) with a theoretically certifiable scheme. Crucially, this occurs without performance degradation (on LM benchmarks) and despite secrets never appearing in the training set.

Apollo: A Posteriori Label-Only Membership Inference Attack Towards Machine Unlearning

Machine Unlearning (MU) aims to update Machine Learning (ML) models following requests to remove training samples and their influences on a trained model efficiently without retraining the original ML model from scratch. While MU itself has been employed to provide privacy protection and regulatory compliance, it can also increase the attack surface of the model. Existing privacy inference attacks towards MU that aim to infer properties of the unlearned set rely on the weaker threat model that assumes the attacker has access to both the unlearned model and the original model, limiting their feasibility toward real-life scenarios. We propose a novel privacy attack, A Posteriori Label-Only Membership Inference Attack towards MU, Apollo, that infers whether a data sample has been unlearned, following a strict threat model where an adversary has access to the label-output of the unlearned model only. We demonstrate that our proposed attack, while requiring less access to the target model compared to previous attacks, can achieve relatively high precision on the membership status of the unlearned samples.

LingoLoop Attack: Trapping MLLMs via Linguistic Context and State Entrapment into Endless Loops

Multimodal Large Language Models (MLLMs) have shown great promise but require substantial computational resources during inference. Attackers can exploit this by inducing excessive output, leading to resource exhaustion and service degradation. Prior energy-latency attacks aim to increase generation time by broadly shifting the output token distribution away from the EOS token, but they neglect the influence of token-level Part-of-Speech (POS) characteristics on EOS and sentence-level structural patterns on output counts, limiting their efficacy. To address this, we propose LingoLoop, an attack designed to induce MLLMs to generate excessively verbose and repetitive sequences. First, we find that the POS tag of a token strongly affects the likelihood of generating an EOS token. Based on this insight, we propose a POS-Aware Delay Mechanism to postpone EOS token generation by adjusting attention weights guided by POS information. Second, we identify that constraining output diversity to induce repetitive loops is effective for sustained generation. We introduce a Generative Path Pruning Mechanism that limits the magnitude of hidden states, encouraging the model to produce persistent loops. Extensive experiments demonstrate LingoLoop can increase generated tokens by up to 30 times and energy consumption by a comparable factor on models like Qwen2.5-VL-3B, consistently driving MLLMs towards their maximum generation limits. These findings expose significant MLLMs' vulnerabilities, posing challenges for their reliable deployment. The code will be released publicly following the paper's acceptance.

SOFT: Selective Data Obfuscation for Protecting LLM Fine-tuning against Membership Inference Attacks

Large language models (LLMs) have achieved remarkable success and are widely adopted for diverse applications. However, fine-tuning these models often involves private or sensitive information, raising critical privacy concerns. In this work, we conduct the first comprehensive study evaluating the vulnerability of fine-tuned LLMs to membership inference attacks (MIAs). Our empirical analysis demonstrates that MIAs exploit the loss reduction during fine-tuning, making them highly effective in revealing membership information. These findings motivate the development of our defense. We propose SOFT (\textbf{S}elective data \textbf{O}bfuscation in LLM \textbf{F}ine-\textbf{T}uning), a novel defense technique that mitigates privacy leakage by leveraging influential data selection with an adjustable parameter to balance utility preservation and privacy protection. Our extensive experiments span six diverse domains and multiple LLM architectures and scales. Results show that SOFT effectively reduces privacy risks while maintaining competitive model performance, offering a practical and scalable solution to safeguard sensitive information in fine-tuned LLMs.

Sharpness-Aware Machine Unlearning

We characterize the effectiveness of Sharpness-aware minimization (SAM) under machine unlearning scheme, where unlearning forget signals interferes with learning retain signals. While previous work prove that SAM improves generalization with noise memorization prevention, we show that SAM abandons such denoising property when fitting the forget set, leading to various test error bounds depending on signal strength. We further characterize the signal surplus of SAM in the order of signal strength, which enables learning from less retain signals to maintain model performance and putting more weight on unlearning the forget set. Empirical studies show that SAM outperforms SGD with relaxed requirement for retain signals and can enhance various unlearning methods either as pretrain or unlearn algorithm. Observing that overfitting can benefit more stringent sample-specific unlearning, we propose Sharp MinMax, which splits the model into two to learn retain signals with SAM and unlearn forget signals with sharpness maximization, achieving best performance. Extensive experiments show that SAM enhances unlearning across varying difficulties measured by data memorization, yielding decreased feature entanglement between retain and forget sets, stronger resistance to membership inference attacks, and a flatter loss landscape.

Network Structures as an Attack Surface: Topology-Based Privacy Leakage in Federated Learning

Federated learning systems increasingly rely on diverse network topologies to address scalability and organizational constraints. While existing privacy research focuses on gradient-based attacks, the privacy implications of network topology knowledge remain critically understudied. We conduct the first comprehensive analysis of topology-based privacy leakage across realistic adversarial knowledge scenarios, demonstrating that adversaries with varying degrees of structural knowledge can infer sensitive data distribution patterns even under strong differential privacy guarantees. Through systematic evaluation of 4,720 attack instances, we analyze six distinct adversarial knowledge scenarios: complete topology knowledge and five partial knowledge configurations reflecting real-world deployment constraints. We propose three complementary attack vectors: communication pattern analysis, parameter magnitude profiling, and structural position correlation, achieving success rates of 84.1%, 65.0%, and 47.2% under complete knowledge conditions. Critically, we find that 80% of realistic partial knowledge scenarios maintain attack effectiveness above security thresholds, with certain partial knowledge configurations achieving performance superior to the baseline complete knowledge scenario. To address these vulnerabilities, we propose and empirically validate structural noise injection as a complementary defense mechanism across 808 configurations, demonstrating up to 51.4% additional attack reduction when properly layered with existing privacy techniques. These results establish that network topology represents a fundamental privacy vulnerability in federated learning systems while providing practical pathways for mitigation through topology-aware defense mechanisms.

Efficiency Robustness of Dynamic Deep Learning Systems

Deep Learning Systems (DLSs) are increasingly deployed in real-time applications, including those in resourceconstrained environments such as mobile and IoT devices. To address efficiency challenges, Dynamic Deep Learning Systems (DDLSs) adapt inference computation based on input complexity, reducing overhead. While this dynamic behavior improves efficiency, such behavior introduces new attack surfaces. In particular, efficiency adversarial attacks exploit these dynamic mechanisms to degrade system performance. This paper systematically explores efficiency robustness of DDLSs, presenting the first comprehensive taxonomy of efficiency attacks. We categorize these attacks based on three dynamic behaviors: (i) attacks on dynamic computations per inference, (ii) attacks on dynamic inference iterations, and (iii) attacks on dynamic output production for downstream tasks. Through an in-depth evaluation, we analyze adversarial strategies that target DDLSs efficiency and identify key challenges in securing these systems. In addition, we investigate existing defense mechanisms, demonstrating their limitations against increasingly popular efficiency attacks and the necessity for novel mitigation strategies to secure future adaptive DDLSs.

Recalling The Forgotten Class Memberships: Unlearned Models Can Be Noisy Labelers to Leak Privacy

Machine Unlearning (MU) technology facilitates the removal of the influence of specific data instances from trained models on request. Despite rapid advancements in MU technology, its vulnerabilities are still underexplored, posing potential risks of privacy breaches through leaks of ostensibly unlearned information. Current limited research on MU attacks requires access to original models containing privacy data, which violates the critical privacy-preserving objective of MU. To address this gap, we initiate an innovative study on recalling the forgotten class memberships from unlearned models (ULMs) without requiring access to the original one. Specifically, we implement a Membership Recall Attack (MRA) framework with a teacher-student knowledge distillation architecture, where ULMs serve as noisy labelers to transfer knowledge to student models. Then, it is translated into a Learning with Noisy Labels (LNL) problem for inferring the correct labels of the forgetting instances. Extensive experiments on state-of-the-art MU methods with multiple real datasets demonstrate that the proposed MRA strategy exhibits high efficacy in recovering class memberships of unlearned instances. As a result, our study and evaluation have established a benchmark for future research on MU vulnerabilities.