Text-to-image (T2I) models have raised increasing safety concerns due to their capacity to generate NSFW and other banned objects. To mitigate these risks, safety filters and concept removal techniques have been introduced to block inappropriate prompts or erase sensitive concepts from the models. However, all the existing defense methods are not well prepared to handle diverse adversarial prompts. In this work, we introduce MacPrompt, a novel black-box and cross-lingual attack that reveals previously overlooked vulnerabilities in T2I safety mechanisms. Unlike existing attacks that rely on synonym substitution or prompt obfuscation, MacPrompt constructs macaronic adversarial prompts by performing cross-lingual character-level recombination of harmful terms, enabling fine-grained control over both semantics and appearance. By leveraging this design, MacPrompt crafts prompts with high semantic similarity to the original harmful inputs (up to 0.96) while bypassing major safety filters (up to 100%). More critically, it achieves attack success rates as high as 92% for sex-related content and 90\% for violence, effectively breaking even state-of-the-art concept removal defenses. These results underscore the pressing need to reassess the robustness of existing T2I safety mechanisms against linguistically diverse and fine-grained adversarial strategies. Warning: This paper includes sensitive examples (e.g., adult, violent, or illegal content). Unsafe images are masked but may still be disturbing.

In recent years, transformer-based models have achieved remarkable success in sensitive domains, including healthcare, finance and personalized services, but their deployment raises significant privacy concerns. Existing secure inference studies have introduced cryptographic techniques such as Homomorphic Encryption (HE) and Secure Multi-Party Computation (MPC). However, these approaches either target isolated model components or incur prohibitive computational and communication overheads, failing to support latency-sensitive or resource-limited environments. In our investigation, we identify substantial redundancy in the nonlinear operations and their alternation with linear layers in deep learning. Motivated by this observation, we propose PCFormer, a universal optimization methodology tailored for sequences of linear and nonlinear computations in the Transformer. PCFormer introduces structure-aware partition and combination techniques specially designed for Multi-Head Attention (MHA) and Feed-Forward Network (FFN). Specifically, we reveal the discrete sources of redundancy in the Softmax and GeLU functions during inference, implementing partitions at the token and channel levels, respectively. Subsequently, these reductions are then combined with the preceding and succeeding linear operations, thereby enhancing both computational and communication efficiency. Experimental results on GLUE benchmarks demonstrate that PCFormer achieves a 1.9× speedup in both computation and communication without compromising accuracy, compared to existing privacy-preserving Transformer frameworks. Furthermore, we demonstrate that PCFormer generalizes effectively to other deep learning architectures involving structured linear-nonlinear compositions under cryptographic constraints.

Despite the integration of safety alignment and external filters, text-to-image (T2I) generative systems are still susceptible to producing harmful content, such as sexual or violent imagery. This raises serious concerns about unintended exposure and potential misuse. Red teaming, which aims to proactively identify diverse prompts that can elicit unsafe outputs from the T2I system, is increasingly recognized as an essential method for assessing and improving safety before real-world deployment. However, existing automated red teaming approaches often treat prompt discovery as an isolated, prompt-level optimization task, which limits their scalability, diversity, and overall effectiveness. To bridge this gap, in this paper, we propose DREAM, a scalable red teaming framework to automatically uncover diverse problematic prompts from a given T2I system. Unlike prior work that optimizes prompts individually, DREAM directly models the probabilistic distribution of the target system's problematic prompts, which enables explicit optimization over both effectiveness and diversity, and allows efficient large-scale sampling after training. To achieve this without direct access to representative training samples, we draw inspiration from energy-based models and reformulate the objective into a simple and tractable form. We further introduce GC-SPSA, an efficient optimization algorithm that provides stable gradient estimates through the long and potentially non-differentiable T2I pipeline. During inference, we also propose a diversity-aware sampling strategy to enhance prompt variety. The effectiveness of DREAM is validated through extensive experiments, demonstrating state-of-the-art performance across a wide range of T2I models and safety filters in terms of both prompt success rate and diversity. Our code is available at https://github.com/AntigoneRandy/DREAM

In recent years, Text-to-Image (T2I) models have garnered significant attention due to their remarkable advancements. However, security concerns have emerged due to their potential to generate inappropriate or Not-Safe-For-Work (NSFW) images. In this paper, inspired by the observation that texts with different semantics can lead to similar human perceptions, we propose an LLM-driven perception-guided jailbreak method, termed PGJ. It is a black-box jailbreak method that requires no specific T2I model (model-free) and generates highly natural attack prompts. Specifically, we propose identifying a safe phrase that is similar in human perception yet inconsistent in text semantics with the target unsafe word and using it as a substitution. The experiments conducted on six open-source models and commercial online services with thousands of prompts have verified the effectiveness of PGJ.

In the last few years, the artifact patterns in fake images synthesized by different generative models have been inconsistent, leading to the failure of previous research that relied on spotting subtle differences between real and fake. In our preliminary experiments, we find that the artifacts in fake images always change with the development of the generative model, while natural images exhibit stable statistical properties. In this paper, we employ natural traces shared only by real images as an additional target for a classifier. Specifically, we introduce a self-supervised feature mapping process for natural trace extraction and develop a transfer learning based on soft contrastive loss to bring them closer to real images and further away from fake ones. This motivates the detector to make decisions based on the proximity of images to the natural traces. To conduct a comprehensive experiment, we built a high-quality and diverse dataset that includes generative models comprising GANs and diffusion models, to evaluate the effectiveness in generalizing unknown forgery techniques and robustness in surviving different transformations. Experimental results show that our proposed method gives 96.2% mAP significantly outperforms the baselines. Extensive experiments conducted on the widely recognized platform Midjourney reveal that our proposed method achieves an accuracy exceeding 78.4%, underscoring its practicality for real-world application deployment.

With the rapid development of the internet, sharing photos through Social Network Platforms (SNPs) has become a new way for people to socialize, which poses serious threats to personal privacy. Recently, a thumbnail-preserving image privacy protection technique has emerged and garnered widespread attention. However, the existing schemes based on this technique often introduce noticeable noise into the protected image, resulting in poor visual quality. Motivated by the observation that a latent vector can be decoupled into the detail and contour components, in this paper, we propose HIPP, a thumbnail-preserving image privacy protection scheme that decouples the detail and contour information contained in the latent vector corresponding to the original image and reconstructs details by generation model. As a result, the generated protected image appears natural and has a thumbnail similar to the original one. Moreover, the protected images can be restored to versions that are indistinguishable from the original images. Experiments on CelebA, Helen, and LSUN datasets show that the SSIM between the restored and original images achieves 0.9899. Furthermore, compared to the previous works, HIPP achieves the lowest runtime and file expansion rate, with values of 0.07 seconds and 1.1046, respectively.

Text-to-image (T2I) diffusion models have achieved impressive image generation quality and are increasingly fine-tuned for personalized applications. However, these models often inherit unsafe behaviors from toxic pretraining data, raising growing safety concerns. While recent safety-driven unlearning methods have made promising progress in suppressing model toxicity, they are found to be fragile to downstream fine-tuning, as we reveal that state-of-the-art methods largely fail to retain their effectiveness even when fine-tuned on entirely benign datasets. To mitigate this problem, in this paper, we propose ResAlign, a safety-driven unlearning framework with enhanced resilience against downstream fine-tuning. By modeling downstream fine-tuning as an implicit optimization problem with a Moreau envelope-based reformulation, ResAlign enables efficient gradient estimation to minimize the recovery of harmful behaviors. Additionally, a meta-learning strategy is proposed to simulate a diverse distribution of fine-tuning scenarios to improve generalization. Extensive experiments across a wide range of datasets, fine-tuning methods, and configurations demonstrate that ResAlign consistently outperforms prior unlearning approaches in retaining safety, while effectively preserving benign generation capability. Our code and pretrained models are publicly available at https://github.com/AntigoneRandy/ResAlign.

Large language models (LLMs) are inherently designed to support multi-turn interactions, which opens up new possibilities for jailbreak attacks that unfold gradually and potentially bypass safety mechanisms more effectively than single-turn attacks. However, current multi-turn jailbreak methods are still in their early stages and suffer from two key limitations. First, they all inherently require inserting sensitive phrases into the context, which makes the dialogue appear suspicious and increases the likelihood of rejection, undermining the effectiveness of the attack. Second, even when harmful content is generated, the response often fails to align with the malicious prompt due to semantic drift, where the conversation slowly moves away from its intended goal. To address these challenges, we propose an analogy-based black-box multi-turn jailbreak framework that constructs fully benign contexts to improve attack success rate while ensuring semantic alignment with the malicious intent. The method first guides the model through safe tasks that mirror the response structure of the malicious prompt, enabling it to internalize the format without exposure to sensitive content. A controlled semantic shift is then introduced in the final turn, substituting benign elements with malicious ones while preserving structural coherence. Experiments on six commercial and open-source LLMs, two benchmark datasets show that our method significantly improves attack performance, achieving an average attack success rate of 93.3\% and outperforming five competitive baselines. Our code is released at https://github.com/MM-WW55/AMA

Text-to-image diffusion models are pushing the boundaries of what generative AI can achieve in our lives. Beyond their ability to generate general images, new personalization techniques have been proposed to customize the pretrained base models for crafting images with specific themes or styles. Such a lightweight solution, enabling AI practitioners and developers to easily build their own personalized models, also poses a new concern regarding whether the personalized models are trained from unauthorized data. A promising solution is to proactively enable data traceability in generative models, where data owners embed external coatings (e.g., image watermarks or backdoor triggers) onto the datasets before releasing. Later the models trained over such datasets will also learn the coatings and unconsciously reproduce them in the generated mimicries, which can be extracted and used as the data usage evidence. However, we identify the existing coatings cannot be effectively learned in personalization tasks, making the corresponding verification less reliable. In this paper, we introduce SIREN, a novel methodology to proactively trace unauthorized data usage in black-box personalized text-to-image diffusion models. Our approach optimizes the coating in a delicate way to be recognized by the model as a feature relevant to the personalization task, thus significantly improving its learnability. We also utilize a human perceptual-aware constraint, a hypersphere classification technique, and a hypothesis-testing-guided verification method to enhance the stealthiness and detection accuracy of the coating. The effectiveness of SIREN is verified through extensive experiments on a diverse set of benchmark datasets, models, and learning algorithms. SIREN is also effective in various real-world scenarios and evaluated against potential countermeasures. Our code is publicly available here.

The pervasive prevalence of DeepFakes poses a profound threat to individual privacy and the stability of society. Believing the synthetic videos of a celebrity and trumping up impersonated forgery videos as authentic are just a few consequences generated by DeepFakes. We investigate current detectors that blindly deploy deep learning techniques that are not effective in capturing subtle clues of forgery when generative models produce remarkably realistic faces. Inspired by the fact that synthetic operations inevitably modify the regions of eyes and mouth to match the target face with the identity or expression of the source face, we conjecture that the continuity of facial movement patterns representing expressions that existed in the veritable faces will be disrupted or completely broken in synthetic faces, making it a potentially formidable indicator for DeepFake detection. To prove this conjecture, we utilize a dual‐branch network to capture the inconsistent patterns of facial movements within eyes and mouth regions separately. Extensive experiments on popular FaceForensics++, Celeb‐DF‐v1, Celeb‐DF‐v2, and DFDC‐Preview datasets have demonstrated not only effectiveness but also the robust capability of our method to outperform the state‐of‐the‐art baselines. Moreover, this work represents greater robustness against adversarial attacks, achieving ASR of 54.8% in the I‐FGSM attack and 43.1% in the PGD attack on the DeepFakes dataset of FaceForensics++, respectively.

Adversarial training (AT) is widely considered the state-of-the-art technique for improving the robustness of deep neural networks (DNNs) against adversarial examples (AEs). Nevertheless, recent studies have revealed that adversarially trained models are prone to unfairness problems. Recent works in this field usually apply class-wise regularization methods to enhance the fairness of AT. However, this paper discovers that these paradigms can be sub-optimal in improving robust fairness. Specifically, we empirically observe that the AEs that are already robust (referred to as “easy AEs” in this paper) are useless and even harmful in improving robust fairness. To this end, we propose the hard adversarial example mining (HAM) technique which concentrates on mining hard AEs while discarding the easy AEs in AT. Specifically, HAM identifies the easy AEs and hard AEs with a fast adversarial attack method. By discarding the easy AEs and reweighting the hard AEs, the robust fairness of the model can be efficiently and effectively improved. Extensive experimental results on four image classification datasets demonstrate the improvement of HAM in robust fairness and training efficiency compared to several state-of-the-art fair adversarial training methods. Our code is available athttps://github.com/yyl-github-1896/HAM.

In recent few years, DeepFakes are posing serve threats and concerns to both individuals and celebrities, as realistic DeepFakes facilitate the spread of disinformation. Model attribution techniques aim at attributing the adopted forgery models of DeepFakes for provenance purposes and providing explainable results to DeepFake forensics. However, the existing model attribution techniques rely on the trace left in the DeepFake creation, which can become futile if such traces were disrupted. Motivated by our observation that certain traces served for model attribution appeared in both the high-frequency and low-frequency domains and play a divergent role in model attribution. In this work, for the first time, we propose a novel training-free evasion attack, TraceEvader, in the most practical non-box setting. Specifically, TraceEvader injects a universal imitated traces learned from wild DeepFakes into the high-frequency component and introduces adversarial blur into the domain of the low-frequency component, where the added distortion confuses the extraction of certain traces for model attribution. The comprehensive evaluation on 4 state-of-the-art (SOTA) model attribution techniques and fake images generated by 8 generative models including generative adversarial networks (GANs) and diffusion models (DMs) demonstrates the effectiveness of our method. Overall, our TraceEvader achieves the highest average attack success rate of 79% and is robust against image transformations and dedicated denoising techniques as well where the average attack success rate is still around 75%. Our TraceEvader confirms the limitations of current model attribution techniques and calls the attention of DeepFake researchers and practitioners for more robust-purpose model attribution techniques.

Generative Adversarial Networks (GANs) have been widely used in various application scenarios. Since the production of a commercial GAN requires substantial computational and human resources, the copyright protection of GANs is urgently needed. This paper presents a novel finger-printing scheme for the Intellectual Property (IP) protection of image-to-image GANs based on a trusted third party. We break through the stealthiness and robustness bottlenecks suffered by previous fingerprinting methods for classification models being naively transferred to GANs. Specifically, we innovatively construct a composite deep learning model from the target GAN and a classifier. Then we generate fingerprint samples from this composite model, and embed them in the classifier for effective ownership verification. This scheme inspires some concrete methodologies to practically protect the modern image-to-image translation GANs. Theoretical analysis proves that these methods can satisfy different security requirements necessary for IP protection. We also conduct extensive experiments to show that our solutions outperform existing strategies.

Model quantization is a compression technique that converts a full-precision model to a more compact low-precision version for better storage. Despite the great success of quantization, recent studies revealed the feasibility of malicious exploiting model quantization via implanting quantization-conditioned backdoors (QCBs). These special backdoors remain dormant in full-precision models but are exposed upon quantization. Unfortunately, existing defenses have limited effects on mitigating QCBs. In this paper, we conduct an in-depth analysis of QCBs. We reveal an intriguing characteristic of QCBs, where activation of backdoor-related neurons on even benign samples enjoy a distribution drift after quantization, although this drift is more significant on poisoned samples. Motivated by this finding, we propose to purify the backdoor-exposed quantized model by aligning its layer-wise activation with its full-precision version. To further exploit the more pronounced activation drifts on poisoned samples, we design an additional module to layer-wisely approximate poisoned activation distribution based on batch normalization statistics of the full-precision model. Extensive experiments are conducted, verifying the effectiveness of our defense. Our code is publicly available.

In recent decades, Generative Adversarial Network (GAN) and its variants have achieved unprecedented success in image synthesis. However, well-trained GANs are under the threat of illegal steal or leakage. The prior studies on remote ownership verification assume a black-box setting where the defender can query the suspicious model with specific inputs, which we identify is not enough for generation tasks. To this end, in this paper, we propose a novel IP protection scheme for GANs where ownership verification can be done by checking outputs only, without choosing the inputs (i.e., box-free setting). Specifically, we make use of the unexploited potential of the discriminator to learn a hypersphere that captures the unique distribution learned by the paired generator. Extensive evaluations on two popular GAN tasks and more than 10 GAN architectures demonstrate our proposed scheme to effectively verify the ownership. Our proposed scheme shown to be immune to popular input-based removal attacks and robust against other existing attacks. The source code and models are available at https://github.com/AbstractTeen/gan_ownership_verification.

Watermarking has been widely adopted for protecting the intellectual property (IP) of Deep Neural Networks (DNN) to defend the unauthorized distribution. Unfortunately, studies have shown that the popular data-poisoning DNN watermarking scheme via tedious model fine-tuning on a poisoned dataset (carefully-crafted sample-label pairs) is not efficient in tackling the tasks on challenging datasets and production-level DNN model protection. To address the aforementioned limitation, in this paper, we propose a plug-and-play watermarking scheme for DNN models by injecting an independent proprietary model into the target model to serve the watermark embedding and ownership verification. In contrast to the prior studies, our proposed method by incorporating a proprietary model is free of target model fine-tuning without involving any parameters update of the target model, thus the fidelity is well preserved and scalable to challenging real tasks. Experimental results on real-world challenging datasets (e.g., ImageNet) and production-level DNN models demonstrated its effectiveness, fidelity w.r.t. the functionality preservation of the target model, robustness against popular watermark removal attacks, and the plug-and-play deployment. The source code and models are available at https://github.com/AntigoneRandy/PTYNet.

Deep neural networks (DNNs) have achieved remarkable success in various tasks (e.g., image classification, speech recognition, and natural language processing (NLP)). However, researchers have demonstrated that DNN-based models are vulnerable to adversarial examples, which cause erroneous predictions by adding imperceptible perturbations into legitimate inputs. Recently, studies have revealed adversarial examples in the text domain, which could effectively evade various DNN-based text analyzers and further bring the threats of the proliferation of disinformation. In this paper, we give a comprehensive survey on the existing studies of adversarial techniques for generating adversarial texts written by both English and Chinese characters and the corresponding defense methods. More importantly, we hope that our work could inspire future studies to develop more robust DNN-based text analyzers against known and unknown adversarial techniques. We classify the existing adversarial techniques for crafting adversarial texts based on the perturbation units, helping to better understand the generation of adversarial texts and build robust models for defense. In presenting the taxonomy of adversarial attacks and defenses in the text domain, we introduce the adversarial techniques from the perspective of different NLP tasks. Finally, we discuss the existing challenges of ad-versarial attacks and defenses in texts and present the future research directions in this emerging and challenging field.

DeepFake is becoming a real risk to society and brings potential threats to both individual privacy and political security due to the DeepFaked multimedia are realistic and convincing. However, the popular DeepFake passive detection is an ex-post forensics countermeasure and failed in blocking the disinformation spreading in advance. To address this limitation, researchers study the proactive defense techniques by adding adversarial noises into the source data to disrupt the DeepFake manipulation. However, the existing studies on proactive DeepFake defense via injecting adversarial noises are not robust, which could be easily bypassed by employing simple image reconstruction revealed in a recent study MagDR. In this paper, we investigate the vulnerability of the existing forgery techniques and propose a novel anti-forgery technique that helps users protect the shared facial images from attackers who are capable of applying the popular forgery techniques. Our proposed method generates perceptual-aware perturbations in an incessant manner which is vastly different from the prior studies by adding adversarial noises that is sparse. Experimental results reveal that our perceptual-aware perturbations are robust to diverse image transformations, especially the competitive evasion technique, MagDR via image reconstruction. Our findings potentially open up a new research direction towards thorough understanding and investigation of perceptual-aware adversarial attack for protecting facial images against DeepFakes in a proactive and robust manner. Code is available at https://github.com/AbstractTeen/AntiForgery.

As deep neural networks (DNNs) play a critical role in various fields, the models themselves hence are becoming an important asset that needs to be protected. To achieve this, various neural network fingerprint methods have been proposed. However, existing fingerprint methods fingerprint the decision boundary by adversarial examples, which is not robust to model modification and adversarial defenses. To fill this gap, we propose a robust fingerprint method MetaFinger, which fingerprints the inner decision area of the model by meta-training, rather than the decision boundary. Specifically, we first generate many shadow models with DNN augmentation as meta-data. Then we optimize some images by meta-training to ensure that only models derived from the protected model can recognize them. To demonstrate the robustness of our fingerprint approach, we evaluate our method against two types of attacks including input modification and model modification. Experiments show that our method achieves 99.34% and 97.69% query accuracy on average, surpassing existing methods over 30%, 25% on CIFAR-10 and Tiny-ImageNet, respectively. Our code is available at https://github.com/kangyangWHU/MetaFinger.

Robustness verification of neural networks (NNs) is a challenging and significant problem, which draws great attention in recent years. Existing researches have shown that bound propagation is a scalable and effective method for robustness verification, and it can be implemented on GPUs and TPUs to get parallelized. However, the bound propagation methods naturally produce weak bound due to linear relaxations on the neurons, which may cause failure in verification. Although tightening techniques for simple ReLU networks have been explored, they are not applicable for NNs with general activation functions such as Sigmoid and Tanh. Improving robustness verification on these NNs is still challenging. In this paper, we propose a Branch-and-Bound (BaB) style method to address this problem. The proposed BaB procedure improves the weak bound by splitting the input domain of neurons into sub-domains and solving the corresponding sub-problems. We propose a generic heuristic function to determine the priority of neuron splitting by scoring the relaxation and impact of neurons. Moreover, we combine bound optimization with the BaB procedure to improve the weak bound. Experimental results demonstrate that the proposed method gains up to 35% improvement compared to the state-of-art CROWN method on Sigmoid and Tanh networks.

Training Deep Neural Networks (DNN) is a time-consuming process and requires a large amount of training data, which motivates studies working on protecting the intellectual property (IP) of DNN models by employing various watermarking techniques. Unfortunately, in recent years, adversaries have been exploiting the vulnerabilities of the employed watermarking techniques to remove the embedded watermarks. In this paper, we investigate and introduce a novel watermark removal attack, called AdvNP, against all the existing four different types of DNN watermarking schemes via input preprocessing by injecting Adversarial Naturalness-aware Perturbations. In contrast to the prior studies, our proposed method is the first work that generalizes all the existing four watermarking schemes well without involving any model modification, which preserves the fidelity of the target model. We conduct the experiments against four state-of-the-art (SOTA) watermarking schemes on two real tasks (e.g., image classification on ImageNet, face recognition on CelebA) across multiple DNN models. Overall, our proposed AdvNP significantly invalidates the watermarks against the four watermarking schemes on two real-world datasets, i.e., 60.9% on the average attack success rate and up to 97% in the worse case. Moreover, our AdvNP could well survive the image denoising techniques and outperforms the baseline in both the fidelity preserving and watermark removal. Furthermore, we introduce two defense methods to enhance the robustness of DNN watermarking against our AdvNP. Our experimental results pose real threats to the existing watermarking schemes and call for more practical and robust watermarking techniques to protect the copyright of pre-trained DNN models. The source code and models are available at ttps://github.com/GitKJ123/AdvNP.

State-of-the-art adversarial attacks in the text domain have shown their power to induce machine learning models to produce abnormal outputs. The samples generated in these attacks have three important attributes: attack ability, transferability, and imperceptibility. However, compared with the other two attributes, the imperceptibility of adversarial examples has not been well investigated. Unlike the pixel-level perturbations in images, adversarial perturbations in the text are usually traceable, reflecting changes in characters, words, or sentences. The generation of imperceptible samples in texts is more difficult than in images. Therefore, how to constrain adversarial perturbations added in the text is a crucial step to construct more natural adversarial texts. Unfortunately, recent studies merely select measurements to constrain the added adversarial perturbations, but none of them explain where these measurements are suitable, which one is better, and how they perform in different kinds of adversarial attacks. In this paper, we fill this gap by comparing the performance of these metrics in various attacks. Furthermore, we propose a stricter constraint for word-level attacks to obtain more imperceptible samples. It is also helpful to enhance existing word-level attacks for adversarial training.

In recent years, DeepFake is becoming a common threat to our society, due to the remarkable progress of generative adversarial networks (GAN) in image synthesis. Unfortunately, existing studies that propose various approaches, in fighting against DeepFake and determining if the facial image is real or fake, is still at an early stage. Obviously, the current DeepFake detection method struggles to catch the rapid progress of GANs, especially in the adversarial scenarios where attackers can evade the detection intentionally, such as adding perturbations to fool the DNN-based detectors. While passive detection simply tells whether the image is fake or real, DeepFake provenance, on the other hand, provides clues for tracking the sources in DeepFake forensics. Thus, the tracked fake images could be blocked immediately by administrators and avoid further spread in social networks.

In recent years, generative adversarial networks (GANs) and its variants have achieved unprecedented success in image synthesis. They are widely adopted in synthesizing facial images which brings potential security concerns to humans as the fakes spread and fuel the misinformation. However, robust detectors of these AI-synthesized fake faces are still in their infancy and are not ready to fully tackle this emerging challenge. In this work, we propose a novel approach, named FakeSpotter, based on monitoring neuron behaviors to spot AI-synthesized fake faces. The studies on neuron coverage and interactions have successfully shown that they can be served as testing criteria for deep learning systems, especially under the settings of being exposed to adversarial attacks. Here, we conjecture that monitoring neuron behavior can also serve as an asset in detecting fake faces since layer-by-layer neuron activation patterns may capture more subtle features that are important for the fake detector. Experimental results on detecting four types of fake faces synthesized with the state-of-the-art GANs and evading four perturbation attacks show the effectiveness and robustness of our approach.

At this moment, GAN-based image generation methods are still imperfect, whose upsampling design has limitations in leaving some certain artifact patterns in the synthesized image. Such artifact patterns can be easily exploited (by recent methods) for difference detection of real and GAN-synthesized images. However, the existing detection methods put much emphasis on the artifact patterns, which can become futile if such artifact patterns were reduced.

Nowadays, digital facial content manipulation has become ubiquitous and realistic with the success of generative adversarial networks (GANs), making face recognition (FR) systems suffer from unprecedented security concerns. In this paper, we investigate and introduce a new type of adversarial attack to evade FR systems by manipulating facial content, called adversarial morphing attack (a.k.a. Amora). In contrast to adversarial noise attack that perturbs pixel intensity values by adding human-imperceptible noise, our proposed adversarial morphing attack works at the semantic level that perturbs pixels spatially in a coherent manner. To tackle the black-box attack problem, we devise a simple yet effective joint dictionary learning pipeline to obtain a proprietary optical flow field for each attack. Our extensive evaluation on two popular FR systems demonstrates the effectiveness of our adversarial morphing attack at various levels of morphing intensity with smiling facial expression manipulations. Both open-set and closed-set experimental results indicate that a novel black-box adversarial attack based on local deformation is possible, and is vastly different from additive noise attacks. The findings of this work potentially pave a new research direction towards a more thorough understanding and investigation of image-based adversarial attacks and defenses.

With the recent advances in voice synthesis, AI-synthesized fake voices are indistinguishable to human ears and widely are applied to produce realistic and natural DeepFakes, exhibiting real threats to our society. However, effective and robust detectors for synthesized fake voices are still in their infancy and are not ready to fully tackle this emerging threat. In this paper, we devise a novel approach, named DeepSonar, based on monitoring neuron behaviors of speaker recognition (SR) system, i.e., a deep neural network (DNN), to discern AI-synthesized fake voices. Layer-wise neuron behaviors provide an important insight to meticulously catch the differences among inputs, which are widely employed for building safety, robust, and interpretable DNNs. In this work, we leverage the power of layer-wise neuron activation patterns with a conjecture that they can capture the subtle differences between real and AI-synthesized fake voices, in providing a cleaner signal to classifiers than raw inputs. Experiments are conducted on three datasets (including commercial products from Google, Baidu, etc) containing both English and Chinese languages to corroborate the high detection rates (98.1% average accuracy) and low false alarm rates (about 2% error rate) of DeepSonar in discerning fake voices. Furthermore, extensive experimental results also demonstrate its robustness against manipulation attacks (e.g., voice conversion and additive real-world noises). Our work further poses a new insight into adopting neuron behaviors for effective and robust AI aided multimedia fakes forensics as an inside-out approach instead of being motivated and swayed by various artifacts introduced in synthesizing fakes.

Recent studies have shown that deep neural networks (DNNs) are susceptible to adversarial attacks even in the black-box settings. However, previous studies on creating black-box based adversarial examples by merely solving the traditional continuous problem, which suffer query efficiency issues. To address the efficiency of querying in black-box attack, we propose a novel attack, called MGAAttack, which is a query-efficient and gradient-free black-box attack without obtaining any knowledge of the target model. In our approach, we leverage the advantages of both transfer-based and scored-based methods, two typical techniques in black-box attack, and solve a discretized problem by using a simple yet effective microbial genetic algorithm (MGA). Experimental results show that our approach dramatically reduces the number of queries on CIFAR-10 and ImageNet and significantly outperforms previous work. In the untargeted attack, we can attack a VGG19 classifier with only 16 queries and give an attack success rate more than 99.90% on ImageNet. Our code is available at https://github.com/kangyangWHU/MGAAttack.

With the unprecedented convenience brought by Apps on mobile devices, we are facing severe security attacks and privacy leakage caused by them since they may stealthily access unclaimed or unneeded permissions for some purposes. Many works strive to discover these malicious apps using program analysis techniques, however, they fail to tell users why an app needs to request the permission from users' perspective. In this paper, we leverage the power of the crowdsourced user reviews to understand why an app requests a permission. We propose a framework, called SmartPI, that automatically identifies functionality-relevant user reviews and infers the permission implication of them, bridging the gap between the functionalities and the actual behaviors of an app. In particular, we extract features from the platform documents to identify functionality-relevant user reviews from noisy crowdsourced user reviews with Natural Language Processing (NLP) techniques. The topic model is further adopted to infer the permission implications of apps from the functionality-relevant user reviews. More than 20,000 apps, 2,653,159 users, and 4,247,769 user reviews are crawled from Google Play as a real-world dataset to evaluate the performance of SmartPI. The experiments results show that the permission usage of apps can be better reflected by user reviews than the claimed descriptions of apps.

Digital password lock has been commonly used on mobile devices as the primary authentication method. Researches have demonstrated that sensors embedded on mobile devices can be employed to infer the password. However, existing works focus on either each single keystroke inference or entire password sequence inference, which are user‐dependent and require huge efforts to collect the ground truth training data. In this paper, we design a novel side‐channel attack system, called Niffler, which leverages the user‐independent features of movements of tapping consecutive buttons to infer unlocking passwords on smartphones. We extract angle features to reflect the changing trends and build a multicategory classifier combining the dynamic time warping algorithm to infer the probability of each movement. We further use the Markov model to model the unlocking process and use the sequences with the highest probabilities as the attack candidates. Moreover, the sensor readings of successful attacks will be further fed back to continually improve the accuracy of the classifier. In our experiments, 100,000 samples collected from 25 participants are used to evaluate the performance of Niffler. The results show that Niffler achieves 70% and 85% accuracy with 10 attempts in user‐independent and user‐dependent environments with few training samples, respectively.

Exploits diagnosis requires great manual effort and desires to be automated as much as possible. In this paper, we investigate how the syntactic format of program inputs, as well as reverse engineering of data structures, could be used to identify overwritten data structures, and propose a binary-level exploit diagnosis approach, deExploit, that is generic to attack types and effective in identifying key attack steps. In details, we design to use a fine-grained dynamic tainting technique to model how the exploit is dynamically processed during program execution, dynamically reverse corresponding data structures of program input and then identify overwritten data structures by detecting the deviation between dynamic processing of exploit and that of benign input. We implement deExploit and perform it to diagnose multiple exploits in the wild. The results show that deExploit works well to diagnose memory corruption exploits.