Kui Ren

Xiaochen Li, Yuke Hu, Weiran Liu et al.

Vertical Federated Learning (FL) is a new paradigm that enables users with non-overlapping attributes of the same data samples to jointly train a model without directly sharing the raw data. Nevertheless, recent works show that it's still not sufficient to prevent privacy leakage from the training process or the trained model. This paper focuses on studying the privacy-preserving tree boosting algorithms under the vertical FL. The existing solutions based on cryptography involve heavy computation and communication overhead and are vulnerable to inference attacks. Although the solution based on Local Differential Privacy (LDP) addresses the above problems, it leads to the low accuracy of the trained model. This paper explores to improve the accuracy of the widely deployed tree boosting algorithms satisfying differential privacy under vertical FL. Specifically, we introduce a framework called OpBoost. Three order-preserving desensitization algorithms satisfying a variant of LDP called distance-based LDP (dLDP) are designed to desensitize the training data. In particular, we optimize the dLDP definition and study efficient sampling distributions to further improve the accuracy and efficiency of the proposed algorithms. The proposed algorithms provide a trade-off between the privacy of pairs with large distance and the utility of desensitized values. Comprehensive evaluations show that OpBoost has a better performance on prediction accuracy of trained models compared with existing LDP approaches on reasonable settings. Our code is open source.

Hongwei Yao, Zheng Li, Kunzhe Huang et al.

With the performance of deep neural networks (DNNs) remarkably improving, DNNs have been widely used in many areas. Consequently, the DNN model has become a valuable asset, and its intellectual property is safeguarded by ownership verification techniques (e.g., DNN fingerprinting). However, the feasibility of the DNN fingerprint removal attack and its potential influence remains an open problem. In this paper, we perform the first comprehensive investigation of DNN fingerprint removal attacks. Generally, the knowledge contained in a DNN model can be categorized into general semantic and fingerprint-specific knowledge. To this end, we propose a min-max bilevel optimization-based DNN fingerprint removal attack named RemovalNet, to evade model ownership verification. The lower-level optimization is designed to remove fingerprint-specific knowledge. While in the upper-level optimization, we distill the victim model's general semantic knowledge to maintain the surrogate model's performance. We conduct extensive experiments to evaluate the fidelity, effectiveness, and efficiency of the RemovalNet against four advanced defense methods on six metrics. The empirical results demonstrate that (1) the RemovalNet is effective. After our DNN fingerprint removal attack, the model distance between the target and surrogate models is x100 times higher than that of the baseline attacks, (2) the RemovalNet is efficient. It uses only 0.2% (400 samples) of the substitute dataset and 1,000 iterations to conduct our attack. Besides, compared with advanced model stealing attacks, the RemovalNet saves nearly 85% of computational resources at most, (3) the RemovalNet achieves high fidelity that the created surrogate model maintains high accuracy after the DNN fingerprint removal process. Our code is available at: https://github.com/grasses/RemovalNet.

Xinyu Zhang, Qingyu Liu, Zhongjie Ba et al.

Federated Learning (FL) is a promising distributed learning approach that enables multiple clients to collaboratively train a shared global model. However, recent studies show that FL is vulnerable to various poisoning attacks, which can degrade the performance of global models or introduce backdoors into them. In this paper, we first conduct a comprehensive study on prior FL attacks and detection methods. The results show that all existing detection methods are only effective against limited and specific attacks. Most detection methods suffer from high false positives, which lead to significant performance degradation, especially in not independent and identically distributed (non-IID) settings. To address these issues, we propose FLTracer, the first FL attack provenance framework to accurately detect various attacks and trace the attack time, objective, type, and poisoned location of updates. Different from existing methodologies that rely solely on cross-client anomaly detection, we propose a Kalman filter-based cross-round detection to identify adversaries by seeking the behavior changes before and after the attack. Thus, this makes it resilient to data heterogeneity and is effective even in non-IID settings. To further improve the accuracy of our detection method, we employ four novel features and capture their anomalies with the joint decisions. Extensive evaluations show that FLTracer achieves an average true positive rate of over $96.88\%$ at an average false positive rate of less than $2.67\%$, significantly outperforming SOTA detection methods. \footnote{Code is available at \url{https://github.com/Eyr3/FLTracer}.}

Shuo Shao, Wenyuan Yang, Hanlin Gu et al.

Federated learning (FL) is a distributed machine learning paradigm allowing multiple clients to collaboratively train a global model without sharing their local data. However, FL entails exposing the model to various participants. This poses a risk of unauthorized model distribution or resale by the malicious client, compromising the intellectual property rights of the FL group. To deter such misbehavior, it is essential to establish a mechanism for verifying the ownership of the model and as well tracing its origin to the leaker among the FL participants. In this paper, we present FedTracker, the first FL model protection framework that provides both ownership verification and traceability. FedTracker adopts a bi-level protection scheme consisting of global watermark mechanism and local fingerprint mechanism. The former authenticates the ownership of the global model, while the latter identifies which client the model is derived from. FedTracker leverages Continual Learning (CL) principles to embed the watermark in a way that preserves the utility of the FL model on both primitive task and watermark task. FedTracker also devises a novel metric to better discriminate different fingerprints. Experimental results show FedTracker is effective in ownership verification, traceability, and maintains good fidelity and robustness against various watermark removal attacks.

Ziqi Yang, Lijin Wang, Da Yang et al.

Neural networks are susceptible to data inference attacks such as the membership inference attack, the adversarial model inversion attack and the attribute inference attack, where the attacker could infer useful information such as the membership, the reconstruction or the sensitive attributes of a data sample from the confidence scores predicted by the target classifier. In this paper, we propose a method, namely PURIFIER, to defend against membership inference attacks. It transforms the confidence score vectors predicted by the target classifier and makes purified confidence scores indistinguishable in individual shape, statistical distribution and prediction label between members and non-members. The experimental results show that PURIFIER helps defend membership inference attacks with high effectiveness and efficiency, outperforming previous defense methods, and also incurs negligible utility loss. Besides, our further experiments show that PURIFIER is also effective in defending adversarial model inversion attacks and attribute inference attacks. For example, the inversion error is raised about 4+ times on the Facescrub530 classifier, and the attribute inference accuracy drops significantly when PURIFIER is deployed in our experiment.

Yukun Chen, Xinyu Zhang, Jialong Tang et al.

While large language models (LLMs) have become pivotal to content safety, current evaluation paradigms primarily focus on detecting explicit harms (e.g., violence or hate speech), neglecting the subtler value dimensions conveyed in digital content. To bridge this gap, we introduce X-Value, a novel Cross-lingual Values Assessment Benchmark designed to evaluate LLMs' ability to assess deep-level values of content from a global perspective. X-Value consists of more than 5,000 QA pairs across 18 languages, systematically organized into 7 core domains grounded in Schwartz's Theory of Basic Human Values and categorized into easy and hard levels for discriminative evaluation. We further propose a unique two-stage annotation framework that first identifies whether an issue falls under global consensus (e.g., human rights) or pluralism (e.g., religion), and subsequently conducts a multi-party evaluation of the latent values embedded within the content. Systematic evaluations on X-Value reveal that current SOTA LLMs exhibit deficiencies in cross-lingual values assessment ($Acc < 77\%$), with significant performance disparities across different languages ($ΔAcc > 20\%$). This work highlights the urgent need to improve the nuanced, values-aware content assessment capability of LLMs. Our X-Value is available at: https://huggingface.co/datasets/Whitolf/X-Value.

Zhibo Wang, Xiaowei Dong, Henry Xue et al.

Prioritizing fairness is of central importance in artificial intelligence (AI) systems, especially for those societal applications, e.g., hiring systems should recommend applicants equally from different demographic groups, and risk assessment systems must eliminate racism in criminal justice. Existing efforts towards the ethical development of AI systems have leveraged data science to mitigate biases in the training set or introduced fairness principles into the training process. For a deployed AI system, however, it may not allow for retraining or tuning in practice. By contrast, we propose a more flexible approach, i.e., fairness-aware adversarial perturbation (FAAP), which learns to perturb input data to blind deployed models on fairness-related features, e.g., gender and ethnicity. The key advantage is that FAAP does not modify deployed models in terms of parameters and structures. To achieve this, we design a discriminator to distinguish fairness-related attributes based on latent representations from deployed models. Meanwhile, a perturbation generator is trained against the discriminator, such that no fairness-related features could be extracted from perturbed inputs. Exhaustive experimental evaluation demonstrates the effectiveness and superior performance of the proposed FAAP. In addition, FAAP is validated on real-world commercial deployments (inaccessible to model parameters), which shows the transferability of FAAP, foreseeing the potential of black-box adaptation.

Xinyu Zhang, Hanbin Hong, Yuan Hong et al.

The language models, especially the basic text classification models, have been shown to be susceptible to textual adversarial attacks such as synonym substitution and word insertion attacks. To defend against such attacks, a growing body of research has been devoted to improving the model robustness. However, providing provable robustness guarantees instead of empirical robustness is still widely unexplored. In this paper, we propose Text-CRS, a generalized certified robustness framework for natural language processing (NLP) based on randomized smoothing. To our best knowledge, existing certified schemes for NLP can only certify the robustness against $\ell_0$ perturbations in synonym substitution attacks. Representing each word-level adversarial operation (i.e., synonym substitution, word reordering, insertion, and deletion) as a combination of permutation and embedding transformation, we propose novel smoothing theorems to derive robustness bounds in both permutation and embedding space against such adversarial operations. To further improve certified accuracy and radius, we consider the numerical relationships between discrete words and select proper noise distributions for the randomized smoothing. Finally, we conduct substantial experiments on multiple language models and datasets. Text-CRS can address all four different word-level adversarial operations and achieve a significant accuracy improvement. We also provide the first benchmark on certified accuracy and radius of four word-level operations, besides outperforming the state-of-the-art certification against synonym substitution attacks.

Yuke Hu, Jian Lou, Jiaqi Liu et al.

Over the past years, Machine Learning-as-a-Service (MLaaS) has received a surging demand for supporting Machine Learning-driven services to offer revolutionized user experience across diverse application areas. MLaaS provides inference service with low inference latency based on an ML model trained using a dataset collected from numerous individual data owners. Recently, for the sake of data owners' privacy and to comply with the "right to be forgotten (RTBF)" as enacted by data protection legislation, many machine unlearning methods have been proposed to remove data owners' data from trained models upon their unlearning requests. However, despite their promising efficiency, almost all existing machine unlearning methods handle unlearning requests independently from inference requests, which unfortunately introduces a new security issue of inference service obsolescence and a privacy vulnerability of undesirable exposure for machine unlearning in MLaaS. In this paper, we propose the ERASER framework for machinE unleaRning in MLaAS via an inferencE seRving-aware approach. ERASER strategically choose appropriate unlearning execution timing to address the inference service obsolescence issue. A novel inference consistency certification mechanism is proposed to avoid the violation of RTBF principle caused by postponed unlearning executions, thereby mitigating the undesirable exposure vulnerability. ERASER offers three groups of design choices to allow for tailor-made variants that best suit the specific environments and preferences of various MLaaS systems. Extensive empirical evaluations across various settings confirm ERASER's effectiveness, e.g., it can effectively save up to 99% of inference latency and 31% of computation overhead over the inference-oblivion baseline.

Zhongjie Ba, Qingyu Liu, Zhenguang Liu et al.

Deepfake technology has given rise to a spectrum of novel and compelling applications. Unfortunately, the widespread proliferation of high-fidelity fake videos has led to pervasive confusion and deception, shattering our faith that seeing is believing. One aspect that has been overlooked so far is that current deepfake detection approaches may easily fall into the trap of overfitting, focusing only on forgery clues within one or a few local regions. Moreover, existing works heavily rely on neural networks to extract forgery features, lacking theoretical constraints guaranteeing that sufficient forgery clues are extracted and superfluous features are eliminated. These deficiencies culminate in unsatisfactory accuracy and limited generalizability in real-life scenarios. In this paper, we try to tackle these challenges through three designs: (1) We present a novel framework to capture broader forgery clues by extracting multiple non-overlapping local representations and fusing them into a global semantic-rich feature. (2) Based on the information bottleneck theory, we derive Local Information Loss to guarantee the orthogonality of local representations while preserving comprehensive task-relevant information. (3) Further, to fuse the local representations and remove task-irrelevant information, we arrive at a Global Information Loss through the theoretical analysis of mutual information. Empirically, our method achieves state-of-the-art performance on five benchmark datasets.Our code is available at \url{https://github.com/QingyuLiu/Exposing-the-Deception}, hoping to inspire researchers.

Xinzhe Huang, Kedong Xiu, Tianhang Zheng et al.

Recent research has focused on exploring the vulnerabilities of Large Language Models (LLMs), aiming to elicit harmful and/or sensitive content from LLMs. However, due to the insufficient research on dual-jailbreaking -- attacks targeting both LLMs and Guardrails, the effectiveness of existing attacks is limited when attempting to bypass safety-aligned LLMs shielded by guardrails. Therefore, in this paper, we propose DualBreach, a target-driven framework for dual-jailbreaking. DualBreach employs a Target-driven Initialization (TDI) strategy to dynamically construct initial prompts, combined with a Multi-Target Optimization (MTO) method that utilizes approximate gradients to jointly adapt the prompts across guardrails and LLMs, which can simultaneously save the number of queries and achieve a high dual-jailbreaking success rate. For black-box guardrails, DualBreach either employs a powerful open-sourced guardrail or imitates the target black-box guardrail by training a proxy model, to incorporate guardrails into the MTO process. We demonstrate the effectiveness of DualBreach in dual-jailbreaking scenarios through extensive evaluation on several widely-used datasets. Experimental results indicate that DualBreach outperforms state-of-the-art methods with fewer queries, achieving significantly higher success rates across all settings. More specifically, DualBreach achieves an average dual-jailbreaking success rate of 93.67% against GPT-4 with Llama-Guard-3 protection, whereas the best success rate achieved by other methods is 88.33%. Moreover, DualBreach only uses an average of 1.77 queries per successful dual-jailbreak, outperforming other state-of-the-art methods. For the purpose of defense, we propose an XGBoost-based ensemble defensive mechanism named EGuard, which integrates the strengths of multiple guardrails, demonstrating superior performance compared with Llama-Guard-3.

Ziping Dong, Chao Shuai, Zhongjie Ba et al.

Invisible Image Watermarking is crucial for ensuring content provenance and accountability in generative AI. While Gen-AI providers are increasingly integrating invisible watermarking systems, the robustness of these schemes against forgery attacks remains poorly characterized. This is critical, as forging traceable watermarks onto illicit content leads to false attribution, potentially harming the reputation and legal standing of Gen-AI service providers who are not responsible for the content. In this work, we propose WMCopier, an effective watermark forgery attack that operates without requiring any prior knowledge of or access to the target watermarking algorithm. Our approach first models the target watermark distribution using an unconditional diffusion model, and then seamlessly embeds the target watermark into a non-watermarked image via a shallow inversion process. We also incorporate an iterative optimization procedure that refines the reconstructed image to further trade off the fidelity and forgery efficiency. Experimental results demonstrate that WMCopier effectively deceives both open-source and closed-source watermark systems (e.g., Amazon's system), achieving a significantly higher success rate than existing methods. Additionally, we evaluate the robustness of forged samples and discuss the potential defenses against our attack.

Kunzhe Huang, Yiming Li, Baoyuan Wu et al.

Recent studies have revealed that deep neural networks (DNNs) are vulnerable to backdoor attacks, where attackers embed hidden backdoors in the DNN model by poisoning a few training samples. The attacked model behaves normally on benign samples, whereas its prediction will be maliciously changed when the backdoor is activated. We reveal that poisoned samples tend to cluster together in the feature space of the attacked DNN model, which is mostly due to the end-to-end supervised training paradigm. Inspired by this observation, we propose a novel backdoor defense via decoupling the original end-to-end training process into three stages. Specifically, we first learn the backbone of a DNN model via \emph{self-supervised learning} based on training samples without their labels. The learned backbone will map samples with the same ground-truth label to similar locations in the feature space. Then, we freeze the parameters of the learned backbone and train the remaining fully connected layers via standard training with all (labeled) training samples. Lastly, to further alleviate side-effects of poisoned samples in the second stage, we remove labels of some `low-credible' samples determined based on the learned model and conduct a \emph{semi-supervised fine-tuning} of the whole model. Extensive experiments on multiple benchmark datasets and DNN models verify that the proposed defense is effective in reducing backdoor threats while preserving high accuracy in predicting benign samples. Our code is available at \url{https://github.com/SCLBD/DBD}.

Zhibo Wang, Hengchang Guo, Zhifei Zhang et al.

Transferability of adversarial examples is of central importance for attacking an unknown model, which facilitates adversarial attacks in more practical scenarios, e.g., black-box attacks. Existing transferable attacks tend to craft adversarial examples by indiscriminately distorting features to degrade prediction accuracy in a source model without aware of intrinsic features of objects in the images. We argue that such brute-force degradation would introduce model-specific local optimum into adversarial examples, thus limiting the transferability. By contrast, we propose the Feature Importance-aware Attack (FIA), which disrupts important object-aware features that dominate model decisions consistently. More specifically, we obtain feature importance by introducing the aggregate gradient, which averages the gradients with respect to feature maps of the source model, computed on a batch of random transforms of the original clean image. The gradients will be highly correlated to objects of interest, and such correlation presents invariance across different models. Besides, the random transforms will preserve intrinsic features of objects and suppress model-specific information. Finally, the feature importance guides to search for adversarial examples towards disrupting critical features, achieving stronger transferability. Extensive experimental evaluation demonstrates the effectiveness and superior performance of the proposed FIA, i.e., improving the success rate by 9.5% against normally trained models and 12.8% against defense models as compared to the state-of-the-art transferable attacks. Code is available at: https://github.com/hcguoO0/FIA

Zhen Cheng, Xinrui Hou, Runhuai Li et al.

We performed the first systematic study of a new attack on Ethereum that steals cryptocurrencies. The attack is due to the unprotected JSON-RPC endpoints existed in Ethereum nodes that could be exploited by attackers to transfer the Ether and ERC20 tokens to attackers-controlled accounts. This study aims to shed light on the attack, including malicious behaviors and profits of attackers. Specifically, we first designed and implemented a honeypot that could capture real attacks in the wild. We then deployed the honeypot and reported results of the collected data in a period of six months. In total, our system captured more than 308 million requests from 1,072 distinct IP addresses. We further grouped attackers into 36 groups with 59 distinct Ethereum accounts. Among them, attackers of 34 groups were stealing the Ether, while other 2 groups were targeting ERC20 tokens. The further behavior analysis showed that attackers were following a three-steps pattern to steal the Ether. Moreover, we observed an interesting type of transaction called zero gas transaction, which has been leveraged by attackers to steal ERC20 tokens. At last, we estimated the overall profits of attackers. To engage the whole community, the dataset of captured attacks is released on https://github.com/zjuicsr/eth-honey.

Tianhang Zheng, Changyou Chen, Junsong Yuan et al.

3D point-cloud recognition with PointNet and its variants has received remarkable progress. A missing ingredient, however, is the ability to automatically evaluate point-wise importance w.r.t.\! classification performance, which is usually reflected by a saliency map. A saliency map is an important tool as it allows one to perform further processes on point-cloud data. In this paper, we propose a novel way of characterizing critical points and segments to build point-cloud saliency maps. Our method assigns each point a score reflecting its contribution to the model-recognition loss. The saliency map explicitly explains which points are the key for model recognition. Furthermore, aggregations of highly-scored points indicate important segments/subsets in a point-cloud. Our motivation for constructing a saliency map is by point dropping, which is a non-differentiable operator. To overcome this issue, we approximate point-dropping with a differentiable procedure of shifting points towards the cloud centroid. Consequently, each saliency score can be efficiently measured by the corresponding gradient of the loss w.r.t the point under the spherical coordinates. Extensive evaluations on several state-of-the-art point-cloud recognition models, including PointNet, PointNet++ and DGCNN, demonstrate the veracity and generality of our proposed saliency map. Code for experiments is released on \url{https://github.com/tianzheng4/PointCloud-Saliency-Maps}.

Tianhang Zheng, Changyou Chen, Kui Ren

Recent work on adversarial attack and defense suggests that PGD is a universal $l_\infty$ first-order attack, and PGD adversarial training can significantly improve network robustness against a wide range of first-order $l_\infty$-bounded attacks, represented as the state-of-the-art defense method. However, an obvious weakness of PGD adversarial training is its highly-computational cost in generating adversarial samples, making it computationally infeasible for large and high-resolution real datasets such as the ImageNet dataset. In addition, recent work also has suggested a simple "close-form" solution to a robust model on MNIST. Therefore, a natural question raised is that is PGD adversarial training really necessary for robust defense? In this paper, we give a negative answer by proposing a training paradigm that is comparable to PGD adversarial training on several standard datasets, while only using noisy-natural samples. Specifically, we reformulate the min-max objective in PGD adversarial training by a problem to minimize the original network loss plus $l_1$ norms of its gradients w.r.t. the inputs. For the $l_1$-norm loss, we propose a computationally-feasible solution by embedding a differentiable soft-quantization layer after the network input layer. We show formally that the soft-quantization layer trained with noisy-natural samples is an alternative approach to minimizing the $l_1$-gradient norms as in PGD adversarial training. Extensive empirical evaluations on standard datasets show that our proposed models are comparable to PGD-adversarially-trained models under PGD and BPDA attacks. Remarkably, our method achieves a 24X speed-up on MNIST while maintaining a comparable defensive ability, and for the first time fine-tunes a robust Imagenet model within only two days. Code is provided on \url{https://github.com/tianzheng4/Noisy-Training-Soft-Quantization}

Jiacheng Du, Jiahui Hu, Zhibo Wang et al.

Federated learning (FL) facilitates collaborative model training among multiple clients without raw data exposure. However, recent studies have shown that clients' private training data can be reconstructed from shared gradients in FL, a vulnerability known as gradient inversion attacks (GIAs). While GIAs have demonstrated effectiveness under \emph{ideal settings and auxiliary assumptions}, their actual efficacy against \emph{practical FL systems} remains under-explored. To address this gap, we conduct a comprehensive study on GIAs in this work. We start with a survey of GIAs that establishes a timeline to trace their evolution and develops a systematization to uncover their inherent threats. By rethinking GIA in practical FL systems, three fundamental aspects influencing GIA's effectiveness are identified: \textit{training setup}, \textit{model}, and \textit{post-processing}. Guided by these aspects, we perform extensive theoretical and empirical evaluations of SOTA GIAs across diverse settings. Our findings highlight that GIA is notably \textit{constrained}, \textit{fragile}, and \textit{easily defensible}. Specifically, GIAs exhibit inherent limitations against practical local training settings. Additionally, their effectiveness is highly sensitive to the trained model, and even simple post-processing techniques applied to gradients can serve as effective defenses. Our work provides crucial insights into the limited threats of GIAs in practical FL systems. By rectifying prior misconceptions, we hope to inspire more accurate and realistic investigations on this topic.

Yiming Li, Shuo Shao, Yu He et al.

The (generative) artificial intelligence (AI) era has profoundly reshaped the meaning and value of data. No longer confined to static content, data now permeates every stage of the AI lifecycle from the training samples that shape model parameters to the prompts and outputs that drive real-world model deployment. This shift renders traditional notions of data protection insufficient, while the boundaries of what needs safeguarding remain poorly defined. Failing to safeguard data in AI systems can inflict societal and individual, underscoring the urgent need to clearly delineate the scope of and rigorously enforce data protection. In this perspective, we propose a four-level taxonomy, including non-usability, privacy preservation, traceability, and deletability, that captures the diverse protection needs arising in modern (generative) AI models and systems. Our framework offers a structured understanding of the trade-offs between data utility and control, spanning the entire AI pipeline, including training datasets, model weights, system prompts, and AI-generated content. We analyze representative technical approaches at each level and reveal regulatory blind spots that leave critical assets exposed. By offering a structured lens to align future AI technologies and governance with trustworthy data practices, we underscore the urgency of rethinking data protection for modern AI techniques and provide timely guidance for developers, researchers, and regulators alike.

Zhixuan Chu, Yan Wang, Longfei Li et al.

Large Language Models (LLMs) have shown impressive performance in natural language tasks, but their outputs can exhibit undesirable attributes or biases. Existing methods for steering LLMs toward desired attributes often assume unbiased representations and rely solely on steering prompts. However, the representations learned from pre-training can introduce semantic biases that influence the steering process, leading to suboptimal results. We propose LLMGuardrail, a novel framework that incorporates causal analysis and adversarial learning to obtain unbiased steering representations in LLMs. LLMGuardrail systematically identifies and blocks the confounding effects of biases, enabling the extraction of unbiased steering representations. Additionally, it includes an explainable component that provides insights into the alignment between the generated output and the desired direction. Experiments demonstrate LLMGuardrail's effectiveness in steering LLMs toward desired attributes while mitigating biases. Our work contributes to the development of safe and reliable LLMs that align with desired attributes.

Tiantian Liu, Hongwei Yao, Feng Lin et al.

Embeddings have become a cornerstone in the functionality of large language models (LLMs) due to their ability to transform text data into rich, dense numerical representations that capture semantic and syntactic properties. These embedding vector databases serve as the long-term memory of LLMs, enabling efficient handling of a wide range of natural language processing tasks. However, the surge in popularity of embedding vector databases in LLMs has been accompanied by significant concerns about privacy leakage. Embedding vector databases are particularly vulnerable to embedding inversion attacks, where adversaries can exploit the embeddings to reverse-engineer and extract sensitive information from the original text data. Existing defense mechanisms have shown limitations, often struggling to balance security with the performance of downstream tasks. To address these challenges, we introduce Eguard, a novel defense mechanism designed to mitigate embedding inversion attacks. Eguard employs a transformer-based projection network and text mutual information optimization to safeguard embeddings while preserving the utility of LLMs. Our approach significantly reduces privacy risks, protecting over 95% of tokens from inversion while maintaining high performance across downstream tasks consistent with original embeddings.

Jiawen Zhang, Kejia Chen, Lipeng He et al.

Large Language Models (LLMs) have showcased remarkable capabilities across various domains. Accompanying the evolving capabilities and expanding deployment scenarios of LLMs, their deployment challenges escalate due to their sheer scale and the advanced yet complex activation designs prevalent in notable model series, such as Llama, Gemma, Mistral. These challenges have become particularly pronounced in resource-constrained deployment scenarios, where mitigating inference bottlenecks is imperative. Among various recent efforts, activation approximation has emerged as a promising avenue for pursuing inference efficiency, sometimes considered indispensable in applications such as private inference. Despite achieving substantial speedups with minimal impact on utility, even appearing sound and practical for real-world deployment, the safety implications of activation approximations remain unclear. In this work, we fill this critical gap in LLM safety by conducting the first systematic safety evaluation of activation approximations. Our safety vetting spans seven state-of-the-art techniques across three popular categories (activation polynomialization, activation sparsification, and activation quantization), revealing consistent safety degradation across ten safety-aligned LLMs. To overcome the hurdle of devising a unified defense accounting for diverse activation approximation methods, we perform an in-depth analysis of their shared error patterns and uncover three key findings. We propose QuadA, a novel safety enhancement method tailored to mitigate the safety compromises introduced by activation approximations. Extensive experiments and ablation studies corroborate QuadA's effectiveness in enhancing the safety capabilities of LLMs after activation approximations.

Qi Zhou, Dongxia Wang, Tianlin Li et al.

Guided image synthesis methods, like SDEdit based on the diffusion model, excel at creating realistic images from user inputs such as stroke paintings. However, existing efforts mainly focus on image quality, often overlooking a key point: the diffusion model represents a data distribution, not individual images. This introduces a low but critical chance of generating images that contradict user intentions, raising ethical concerns. For example, a user inputting a stroke painting with female characteristics might, with some probability, get male faces from SDEdit. To expose this potential vulnerability, we aim to build an adversarial attack forcing SDEdit to generate a specific data distribution aligned with a specified attribute (e.g., female), without changing the input's attribute characteristics. We propose the Targeted Attribute Generative Attack (TAGA), using an attribute-aware objective function and optimizing the adversarial noise added to the input stroke painting. Empirical studies reveal that traditional adversarial noise struggles with TAGA, while natural perturbations like exposure and motion blur easily alter generated images' attributes. To execute effective attacks, we introduce FoolSDEdit: We design a joint adversarial exposure and blur attack, adding exposure and motion blur to the stroke painting and optimizing them together. We optimize the execution strategy of various perturbations, framing it as a network architecture search problem. We create the SuperPert, a graph representing diverse execution strategies for different perturbations. After training, we obtain the optimized execution strategy for effective TAGA against SDEdit. Comprehensive experiments on two datasets show our method compelling SDEdit to generate a targeted attribute-aware data distribution, significantly outperforming baselines.

Zhongjie Ba, Yitao Zhang, Peng Cheng et al.

Watermarking plays a key role in the provenance and detection of AI-generated content. While existing methods prioritize robustness against real-world distortions (e.g., JPEG compression and noise addition), we reveal a fundamental tradeoff: such robust watermarks inherently improve the redundancy of detectable patterns encoded into images, creating exploitable information leakage. To leverage this, we propose an attack framework that extracts leakage of watermark patterns through multi-channel feature learning using a pre-trained vision model. Unlike prior works requiring massive data or detector access, our method achieves both forgery and detection evasion with a single watermarked image. Extensive experiments demonstrate that our method achieves a 60\% success rate gain in detection evasion and 51\% improvement in forgery accuracy compared to state-of-the-art methods while maintaining visual fidelity. Our work exposes the robustness-stealthiness paradox: current "robust" watermarks sacrifice security for distortion resistance, providing insights for future watermark design.

Sifan Wu, Zhenguang Liu, Beibei Zhang et al.

Human motion copy is an intriguing yet challenging task in artificial intelligence and computer vision, which strives to generate a fake video of a target person performing the motion of a source person. The problem is inherently challenging due to the subtle human-body texture details to be generated and the temporal consistency to be considered. Existing approaches typically adopt a conventional GAN with an L1 or L2 loss to produce the target fake video, which intrinsically necessitates a large number of training samples that are challenging to acquire. Meanwhile, current methods still have difficulties in attaining realistic image details and temporal consistency, which unfortunately can be easily perceived by human observers. Motivated by this, we try to tackle the issues from three aspects: (1) We constrain pose-to-appearance generation with a perceptual loss and a theoretically motivated Gromov-Wasserstein loss to bridge the gap between pose and appearance. (2) We present an episodic memory module in the pose-to-appearance generation to propel continuous learning that helps the model learn from its past poor generations. We also utilize geometrical cues of the face to optimize facial details and refine each key body part with a dedicated local GAN. (3) We advocate generating the foreground in a sequence-to-sequence manner rather than a single-frame manner, explicitly enforcing temporal inconsistency. Empirical results on five datasets, iPER, ComplexMotion, SoloDance, Fish, and Mouse datasets, demonstrate that our method is capable of generating realistic target videos while precisely copying motion from a source video. Our method significantly outperforms state-of-the-art approaches and gains 7.2% and 12.4% improvements in PSNR and FID respectively.

Zhibo Wang, Zhiwei Chang, Jiahui Hu et al.

Federated Learning (FL) exhibits privacy vulnerabilities under gradient inversion attacks (GIAs), which can extract private information from individual gradients. To enhance privacy, FL incorporates Secure Aggregation (SA) to prevent the server from obtaining individual gradients, thus effectively resisting GIAs. In this paper, we propose a stealthy label inference attack to bypass SA and recover individual clients' private labels. Specifically, we conduct a theoretical analysis of label inference from the aggregated gradients that are exclusively obtained after implementing SA. The analysis results reveal that the inputs (embeddings) and outputs (logits) of the final fully connected layer (FCL) contribute to gradient disaggregation and label restoration. To preset the embeddings and logits of FCL, we craft a fishing model by solely modifying the parameters of a single batch normalization (BN) layer in the original model. Distributing client-specific fishing models, the server can derive the individual gradients regarding the bias of FCL by resolving a linear system with expected embeddings and the aggregated gradients as coefficients. Then the labels of each client can be precisely computed based on preset logits and gradients of FCL's bias. Extensive experiments show that our attack achieves large-scale label recovery with 100\% accuracy on various datasets and model architectures.

Zhuo Chen, Lei Wu, Jing Cheng et al.

Mobile apps are extensively involved in cyber-crimes. Some apps are malware which compromise users' devices, while some others may lead to privacy leakage. Apart from them, there also exist apps which directly make profit from victims through deceiving, threatening or other criminal actions. We name these apps as CULPRITWARE. They have become emerging threats in recent years. However, the characteristics and the ecosystem of CULPRITWARE remain mysterious. This paper takes the first step towards systematically studying CULPRITWARE and its ecosystem. Specifically, we first characterize CULPRITWARE by categorizing and comparing them with benign apps and malware. The result shows that CULPRITWARE have unique features, e.g., the usage of app generators (25.27%) deviates from that of benign apps (5.08%) and malware (0.43%). Such a discrepancy can be used to distinguish CULPRITWARE from benign apps and malware. Then we understand the structure of the ecosystem by revealing the four participating entities (i.e., developer, agent, operator and reaper) and the workflow. After that, we further reveal the characteristics of the ecosystem by studying the participating entities. Our investigation shows that the majority of CULPRITWARE (at least 52.08%) are propagated through social media rather than the official app markets, and most CULPRITWARE (96%) indirectly rely on the covert fourth-party payment services to transfer the profits. Our findings shed light on the ecosystem, and can facilitate the community and law enforcement authorities to mitigate the threats. We will release the source code of our tools to engage the community.

Dingding Wang, Muhui Jiang, Rui Chang et al.

Embedded devices are becoming popular. Meanwhile, researchers are actively working on improving the security of embedded devices. However, previous work ignores the insecurity caused by a special category of devices, i.e., the End-of-Life (EoL in short) devices. Once a product becomes End-of-Life, vendors tend to no longer maintain its firmware or software, including providing bug fixes and security patches. This makes EoL devices susceptible to attacks. For instance, a report showed that an EoL model with thousands of active devices was exploited to redirect web traffic for malicious purposes. In this paper, we conduct the first measurement study to shed light on the (in)security of EoL devices. To this end, our study performs two types of analysis, including the aliveness analysis and the vulnerability analysis. The first one aims to detect the scale of EoL devices that are still alive. The second one is to evaluate the vulnerabilities existing in (active) EoL devices. We have applied our approach to a large number of EoL models from three vendors (i.e., D-Link, Tp-Link, and Netgear) and detect the alive devices in a time period of ten months. Our study reveals some worrisome facts that were unknown by the community. For instance, there exist more than 2 million active EoL devices. Nearly 300,000 of them are still alive even after five years since they became EoL. Although vendors may release security patches after the EoL date, however, the process is ad hoc and incomplete. As a result, more than 1 million active EoL devices are vulnerable, and nearly half of them are threatened by high-risk vulnerabilities. Attackers can achieve a minimum of 2.79 Tbps DDoS attack by compromising a large number of active EoL devices. We believe these facts pose a clear call for more attention to deal with the security issues of EoL devices.

Muhui Jiang, Lin Ma, Yajin Zhou et al.

Dynamic analysis based on the full-system emulator QEMU is widely used for various purposes. However, it is challenging to run firmware images of embedded devices in QEMU, especially the process to boot the Linux kernel (we call this process rehosting the Linux kernel.) That's because embedded devices usually use different system-on-chips (SoCs) from multiple vendors and only a limited number of SoCs are currently supported in QEMU. In this work, we propose a technique called peripheral transplantation. The main idea is to transplant the device drivers of designated peripherals into the Linux kernel. By doing so, it can replace the peripherals in the kernel that are currently unsupported in QEMU with supported ones, thus making the Linux kernel rehostable. After that, various applications can be built upon. We implemented this technique inside a prototype system called ECMO and applied it to 815 firmware images, which consist of 20 kernel versions, 37 device models, and 24 vendors. The result shows that ECMO can successfully transplant peripherals for all the 815 Linux kernels. Among them,710 kernels can be successfully rehosted, i.e., launching a user-space shell (87.1% success rate). The failed cases are mainly because the root file system format (ramfs) is not supported by the kernel. We further build three applications, i.e., kernel crash analysis, rootkit forensic analysis, and kernel fuzzing, based on the rehosted kernels to demonstrate the usage scenarios of ECMO

Lin Ma, Jinyan Xu, Jiadong Sun et al.

Selective data protection is a promising technique to defend against the data leakage attack. In this paper, we revisit technical challenges that were neglected when applying this protection to real applications. These challenges include the secure input channel, granularity conflict, and sensitivity conflict. We summarize the causes of them and propose corresponding solutions. Then we design and implement a prototype system for selective data protection and evaluate the overhead using the RISC-V Spike simulator. The evaluation demonstrates the efficiency (less than 3% runtime overhead with optimizations) and the security guarantees provided by our system.

Siwei Wu, Dabao Wang, Jianting He et al.

The rapid growth of Decentralized Finance (DeFi) boosts the Ethereum ecosystem. At the same time, attacks towards DeFi applications (apps) are increasing. However, to the best of our knowledge, existing smart contract vulnerability detection tools cannot be directly used to detect DeFi attacks. That's because they lack the capability to recover and understand high-level DeFi semantics, e.g., a user trades a token pair X and Y in a Decentralized EXchange (DEX). In this work, we focus on the detection of two types of new attacks on DeFi apps, including direct and indirect price manipulation attacks. The former one means that an attacker directly manipulates the token price in DEX by performing an unwanted trade in the same DEX by attacking the vulnerable DeFi app. The latter one means that an attacker indirectly manipulates the token price of the vulnerable DeFi app (e.g., a lending app). To this end, we propose a platform-independent way to recover high-level DeFi semantics by first constructing the cash flow tree from raw Ethereum transactions and then lifting the low-level semantics to high-level ones, including token trade, liquidity mining, and liquidity cancel. Finally, we detect price manipulation attacks using the patterns expressed with the recovered DeFi semantics. We have implemented a prototype named \tool{} and applied it to more than 350 million transactions. It successfully detected 432 real-world attacks in the wild. We confirm that they belong to four known security incidents and five zero-day ones. We reported our findings. Two CVEs have been assigned. We further performed an attack analysis to reveal the root cause of the vulnerability, the attack footprint, and the impact of the attack. Our work urges the need to secure the DeFi ecosystem.

Muhammad Saad, Zhan Qin, Kui Ren et al.

Blockchain applications that rely on the Proof-of-Work (PoW) have increasingly become energy inefficient with a staggering carbon footprint. In contrast, energy-efficient alternative consensus protocols such as Proof-of-Stake (PoS) may cause centralization and unfairness in the blockchain system. To address these challenges, we propose a modular version of PoS-based blockchain systems called epos that resists the centralization of network resources by extending mining opportunities to a wider set of stakeholders. Moreover, epos leverages the in-built system operations to promote fair mining practices by penalizing malicious entities. We validate epos's achievable objectives through theoretical analysis and simulations. Our results show that epos ensures fairness and decentralization, and can be applied to existing blockchain applications.

Weilin Li, Kui Ren, Donsub Rim

This work characterizes the range of the single-quadrant approximate discrete Radon transform (ADRT) of square images. The characterization follows from a set of linear constraints on the codomain. We show that for data satisfying these constraints, the exact and fast inversion formula [Rim, Appl. Math. Lett. 102 106159, 2020] yields a square image in a stable manner. The range characterization is obtained by first showing that the ADRT is a bijection between images supported on infinite half-strips, then identifying the linear subspaces that stay finitely supported under the inversion formula.

Sekhar Rajendran, Zhi Sun, Feng Lin et al.

In Internet of Things, where billions of devices with limited resources are communicating with each other, security has become a major stumbling block affecting the progress of this technology. Existing authentication schemes-based on digital signatures have overhead costs associated with them in terms of computation time, battery power, bandwidth, memory, and related hardware costs. Radio frequency fingerprint (RFF), utilizing the unique device-based information, can be a promising solution for IoT. However, traditional RFFs have become obsolete because of low reliability and reduced user capability. Our proposed solution, Metasurface RF-Fingerprinting Injection (MeRFFI), is to inject a carefully-designed radio frequency fingerprint into the wireless physical layer that can increase the security of a stationary IoT device with minimal overhead. The injection of fingerprint is implemented using a low cost metasurface developed and fabricated in our lab, which is designed to make small but detectable perturbations in the specific frequency band in which the IoT devices are communicating. We have conducted comprehensive system evaluations including distance, orientation, multiple channels where the feasibility, effectiveness, and reliability of these fingerprints are validated. The proposed MeRFFI system can be easily integrated into the existing authentication schemes. The security vulnerabilities are analyzed for some of the most threatening wireless physical layer-based attacks.

Lei Wu, Siwei Wu, Yajin Zhou et al.

As one of the representative blockchain platforms, Ethereum has attracted lots of attacks. Due to the existed financial loss, there is a pressing need to perform timely investigation and detect more attack instances. Though multiple systems have been proposed, they suffer from the scalability issue due to the following reasons. First, the tight coupling between malicious contract detection and blockchain data importing makes them infeasible to repeatedly detect different attacks. Second, the coarse-grained archive data makes them inefficient to replay transactions. Third, the separation between malicious contract detection and runtime state recovery consumes lots of storage. In this paper, we present the design of a scalable attack detection framework on Ethereum. It overcomes the scalability issue by saving the Ethereum state into a database and providing an efficient way to locate suspicious transactions. The saved state is fine-grained to support the replay of arbitrary transactions. The state is well-designed to avoid saving unnecessary state to optimize the storage consumption. We implement a prototype named EthScope and solve three technical challenges, i.e., incomplete Ethereum state, scalability, and extensibility. The performance evaluation shows that our system can solve the scalability issue, i.e., efficiently performing a large-scale analysis on billions of transactions, and a speedup of around 2,300x when replaying transactions. It also has lower storage consumption compared with existing systems. The result with three different types of information as inputs shows that our system can help an analyst understand attack behaviors and further detect more attacks. To engage the community, we will release our system and the dataset of detected attacks.

Tianhang Zheng, Sheng Liu, Changyou Chen et al.

Skeleton-based action recognition has attracted increasing attention due to its strong adaptability to dynamic circumstances and potential for broad applications such as autonomous and anonymous surveillance. With the help of deep learning techniques, it has also witnessed substantial progress and currently achieved around 90\% accuracy in benign environment. On the other hand, research on the vulnerability of skeleton-based action recognition under different adversarial settings remains scant, which may raise security concerns about deploying such techniques into real-world systems. However, filling this research gap is challenging due to the unique physical constraints of skeletons and human actions. In this paper, we attempt to conduct a thorough study towards understanding the adversarial vulnerability of skeleton-based action recognition. We first formulate generation of adversarial skeleton actions as a constrained optimization problem by representing or approximating the physiological and physical constraints with mathematical formulations. Since the primal optimization problem with equality constraints is intractable, we propose to solve it by optimizing its unconstrained dual problem using ADMM. We then specify an efficient plug-in defense, inspired by recent theories and empirical observations, against the adversarial skeleton actions. Extensive evaluations demonstrate the effectiveness of the attack and defense method under different settings.

Jianwei Liu, Jinsong Han, Lei Yang et al.

Recent years have witnessed the bloom development of the human-centered wireless sensing applications, in which some human information, such as the user's identity and motions, can be retrieved through analyzing the signal distortion caused by the target person. However, the openness of wireless transmission raises increasing concerns on user privacy, since either the human identity or human motion is sensitive in certain scenarios, including personal residence, laboratory, and office. Researchers have reported that commodity WiFi signals can be abused to identify users. To dispel this threat, in this paper we propose a privacy-preserving framework to effectively hide the information of user behaviors in wireless signals while retaining the ability of user authentication. The core of our framework is a novel Siamese network-based deep model, namely RFBP-Net. In this way, wireless sensing reveals user information moderately. We conduct extensive experiments on both the real WiFi and RFID system and open datasets. The experiment results show that RFBP-Net is able to significantly reduce the activity recognition accuracy, i.e., 70% reduction in the RFID system and 80% reduction in the WiFi system, with a slight penalty in the user authentication accuracy, i.e., only 5% and 1% decrease in the RFID and WiFi system, respectively.

Yutian Yang, Songbo Zhu, Wenbo Shen et al.

Code reuse attacks are still big threats to software and system security. Control flow integrity is a promising technique to defend against such attacks. However, its effectiveness has been weakened due to the inaccurate control flow graph and practical strategy to trade security for performance. In recent years, CPU vendors have integrated hardware features as countermeasures. For instance, ARM Pointer Authentication (PA in short) was introduced in ARMV8-A architecture. It can efficiently generate an authentication code for an address, which is encoded in the unused bits of the address. When the address is de-referenced, the authentication code is checked to ensure its integrity. Though there exist systems that adopt PA to harden user programs, how to effectively use PA to protect OS kernels is still an open research question. In this paper, we shed lights on how to leverage PA to protect control flows, including function pointers and return addresses, of Linux kernel. Specifically, to protect function pointers, we embed authentication code into them, track their propagation and verify their values when loading from memory or branching to targets. To further defend against the pointer substitution attack, we use the function pointer address as its context, and take a clean design to propagate the address by piggybacking it into the pointer value. We have implemented a prototype system with LLVM to identify function pointers, add authentication code and verify function pointers by emitting new machine instructions. We applied this system to Linux kernel, and solved numerous practical issues, e.g., function pointer comparison and arithmetic operations. The security analysis shows that our system can protect all function pointers and return addresses in Linux kernel.

Hengtong Zhang, Tianhang Zheng, Jing Gao et al.

Knowledge graph embedding (KGE) is a technique for learning continuous embeddings for entities and relations in the knowledge graph.Due to its benefit to a variety of downstream tasks such as knowledge graph completion, question answering and recommendation, KGE has gained significant attention recently. Despite its effectiveness in a benign environment, KGE' robustness to adversarial attacks is not well-studied. Existing attack methods on graph data cannot be directly applied to attack the embeddings of knowledge graph due to its heterogeneity. To fill this gap, we propose a collection of data poisoning attack strategies, which can effectively manipulate the plausibility of arbitrary targeted facts in a knowledge graph by adding or deleting facts on the graph. The effectiveness and efficiency of the proposed attack strategies are verified by extensive evaluations on two widely-used benchmarks.

Fei Wang, Jinsong Han, Feng Lin et al.

Wi-Fi signals-based person identification attracts increasing attention in the booming Internet-of-Things era mainly due to its pervasiveness and passiveness. Most previous work applies gaits extracted from WiFi distortions caused by the person walking to achieve the identification. However, to extract useful gait, a person must walk along a pre-defined path for several meters, which requires user high collaboration and increases identification time overhead, thus limiting use scenarios. Moreover, gait based work has severe shortcoming in identification performance, especially when the user volume is large. In order to eliminate the above limitations, in this paper, we present an operation-free person identification system, namely WiPIN, that requires least user collaboration and achieves good performance. WiPIN is based on an entirely new insight that Wi-Fi signals would carry person body information when propagating through the body, which is potentially discriminated for person identification. Then we demonstrate the feasibility on commodity off-the-shelf Wi-Fi devices by well-designed signal pre-processing, feature extraction, and identity matching algorithms. Results show that WiPIN achieves 92% identification accuracy over 30 users, high robustness to various experimental settings, and low identifying time overhead, i.e., less than 300ms.