Empirical results in Section 3.1 in the main paper show that simply removing ReLUs lead to improved transferability. In this section, we try freezing all learnable parameters in the unmodified sub-net h during fine-tuning and a similar observation about the initial improvement of transferability can still be decrease made and (see finally Figure the 5). Classification loss of these modified VGG-19 models on the benign CIFAR-10 test set is also reported, in Figure 6. On ImageNet, it is evaluated on the 50000official validation images. As mentioned in the main paper, many recent successes in improving adversarial transferability benefit from maximizing intermediate level distortions rather than the final prediction losses [8, 3, 2] of DNNs.

Adversarial prompts (or say, adversarial examples) generated using gradient-based methods exhibit outstanding performance in performing automatic jailbreak attacks against safety-aligned LLMs. Nevertheless, due to the discrete nature of texts, the input gradient of LLMs struggles to precisely reflect the magnitude of loss change that results from token replacements in the prompt, leading to limited attack success rates against safety-aligned LLMs, even in the setting. In this paper, we explore a new perspective on this problem, suggesting that it can be alleviated by leveraging innovations inspired in transfer-based attacks that were originally proposed for attacking image classification models. For the first time, we appropriate the ideologies of effective methods among these transfer-based attacks,, Skip Gradient Method and Intermediate Level Attack, into gradient-based adversarial prompt generation and achieve significant performance gains without introducing obvious computational cost. Meanwhile, by discussing mechanisms behind the gains, new insights are drawn, and proper combinations of these methods are also developed. Our empirical results show that 87% of the query-specific adversarial suffixes generated by the developed combination can induce Llama-2-7B-Chat to produce the output that exactly matches the target string on AdvBench. This match rate is 33% higher than that of a very strong baseline known as GCG, demonstrating advanced discrete optimization for adversarial prompt generation against LLMs.

Neural Information Processing Systems http://nips.cc/

Transfer-based attacks are a practical method of black-box adversarial attacks, in which the attacker aims to craft adversarial examples from a source (surrogate) model that is transferable to the target model. A wide range of empirical works has tried to explain the transferability of adversarial examples from different angles. However, these works only provide ad hoc explanations without quantitative analyses. The theory behind transfer-based attacks remains a mystery.This paper studies transfer-based attacks under a unified theoretical framework. We propose an explanatory model, called the manifold attack model, that formalizes popular beliefs and explains the existing empirical results. Our model explains why adversarial examples are transferable even when the source model is inaccurate. Moreover, our model implies that the existence of transferable adversarial examples depends on the "curvature" of the data manifold, which quantitatively explains why the success rates of transfer-based attacks are hard to improve. We also discuss the expressive power and the possible extensions of our model in general applications.

The prosperous development of Artificial Intelligence-Generated Content (AIGC) has brought people's anxiety about the spread of false information on social media. Designing detectors for filtering is an effective defense method, but most detectors will be compromised by adversarial samples. Currently, most studies exposing AIGC security issues assume information on model structure and data distribution. In real applications, attackers query and interfere with models that provide services in the form of application programming interfaces (APIs), which constitutes the black-box decision-based attack paradigm. However, to the best of our knowledge, decision-based attacks on AIGC detectors remain unexplored. In this study, we propose \textbf{FBA$^2$D}: a frequency-based black-box attack method for AIGC detection to fill the research gap. Motivated by frequency-domain discrepancies between generated and real images, we develop a decision-based attack that leverages the Discrete Cosine Transform (DCT) for fine-grained spectral partitioning and selects frequency bands as query subspaces, improving both query efficiency and image quality. Moreover, attacks on AIGC detectors should mitigate initialization failures, preserve image quality, and operate under strict query budgets. To address these issues, we adopt an ``adversarial example soup'' method, averaging candidates from successive surrogate iterations and using the result as the initialization to accelerate the query-based attack. The empirical study on the Synthetic LSUN dataset and GenImage dataset demonstrate the effectiveness of our prosed method. This study shows the urgency of addressing practical AIGC security problems.

Neural Information Processing Systems http://nips.cc/

Adversarial prompts (or say, adversarial examples) generated using gradient-based methods exhibit outstanding performance in performing automatic jailbreak attacks against safety-aligned LLMs. Nevertheless, due to the discrete nature of texts, the input gradient of LLMs struggles to precisely reflect the magnitude of loss change that results from token replacements in the prompt, leading to limited attack success rates against safety-aligned LLMs, even in the white-box setting. In this paper, we explore a new perspective on this problem, suggesting that it can be alleviated by leveraging innovations inspired in transfer-based attacks that were originally proposed for attacking black-box image classification models. For the first time, we appropriate the ideologies of effective methods among these transfer-based attacks, i.e., Skip Gradient Method and Intermediate Level Attack, into gradient-based adversarial prompt generation and achieve significant performance gains without introducing obvious computational cost. Meanwhile, by discussing mechanisms behind the gains, new insights are drawn, and proper combinations of these methods are also developed.

Much effort has been made to explain and improve the success of transfer-based attacks (TBA) on black-box computer vision models. This work provides the first attempt at a priori prediction of attack success by identifying the presence of vulnerable features within target models. Recent work by Chen and Liu (2024) proposed the manifold attack model, a unifying framework proposing that successful TBA exist in a common manifold space. Our work experimentally tests the common manifold space hypothesis by a new methodology: first, projecting feature vectors from surrogate and target feature extractors trained on ImageNet onto the same low-dimensional manifold; second, quantifying any observed structure similarities on the manifold; and finally, by relating these observed similarities to the success of the TBA. We find that shared feature representation moderately correlates with increased success of TBA (\r{ho}= 0.56). This method may be used to predict whether an attack will transfer without information of the model weights, training, architecture or details of the attack. The results confirm the presence of shared feature representations between two feature extractors of different sizes and complexities, and demonstrate the utility of datasets from different target domains as test signals for interpreting black-box feature representations.

Skeletal sequences, as well-structured representations of human behaviors, are crucial in Human Activity Recognition (HAR). The transferability of adversarial skeletal sequences enables attacks in real-world HAR scenarios, such as autonomous driving, intelligent surveillance, and human-computer interactions. However, existing Skeleton-based HAR (S-HAR) attacks exhibit weak adversarial transferability and, therefore, cannot be considered true transfer-based S-HAR attacks. More importantly, the reason for this failure remains unclear. In this paper, we study this phenomenon through the lens of loss surface, and find that its sharpness contributes to the poor transferability in S-HAR. Inspired by this observation, we assume and empirically validate that smoothening the rugged loss landscape could potentially improve adversarial transferability in S-HAR. To this end, we propose the first Transfer-based Attack on Skeletal Action Recognition, TASAR. TASAR explores the smoothed model posterior without re-training the pre-trained surrogates, which is achieved by a new post-train Dual Bayesian optimization strategy. Furthermore, unlike previous transfer-based attacks that treat each frame independently and overlook temporal coherence within sequences, TASAR incorporates motion dynamics into the Bayesian attack gradient, effectively disrupting the spatial-temporal coherence of S-HARs. To exhaustively evaluate the effectiveness of existing methods and our method, we build the first large-scale robust S-HAR benchmark, comprising 7 S-HAR models, 10 attack methods, 3 S-HAR datasets and 2 defense models. Extensive results demonstrate the superiority of TASAR. Our benchmark enables easy comparisons for future studies, with the code available in the supplementary material.

Split learning enables collaborative deep learning model training while preserving data privacy and model security by avoiding direct sharing of raw data and model details (i.e., sever and clients only hold partial sub-networks and exchange intermediate computations). However, existing research has mainly focused on examining its reliability for privacy protection, with little investigation into model security. Specifically, by exploring full models, attackers can launch adversarial attacks, and split learning can mitigate this severe threat by only disclosing part of models to untrusted servers.This paper aims to evaluate the robustness of split learning against adversarial attacks, particularly in the most challenging setting where untrusted servers only have access to the intermediate layers of the model.Existing adversarial attacks mostly focus on the centralized setting instead of the collaborative setting, thus, to better evaluate the robustness of split learning, we develop a tailored attack called SPADV, which comprises two stages: 1) shadow model training that addresses the issue of lacking part of the model and 2) local adversarial attack that produces adversarial examples to evaluate.The first stage only requires a few unlabeled non-IID data, and, in the second stage, SPADV perturbs the intermediate output of natural samples to craft the adversarial ones. The overall cost of the proposed attack process is relatively low, yet the empirical attack effectiveness is significantly high, demonstrating the surprising vulnerability of split learning to adversarial attacks.