Transfer adversarial attacks raise critical security concerns in real-world, black-box scenarios. However, the actual progress of this field is difficult to assess due to two common limitations in existing evaluations. First, different methods are often not systematically and fairly evaluated in a one-to-one comparison. Second, only transferability is evaluated but another key attack property, stealthiness, is largely overlooked. In this work, we design good practices to address these limitations, and we present the first comprehensive evaluation of transfer attacks, covering 23 representative attacks against 9 defenses on ImageNet. In particular, we propose to categorize existing attacks into five categories, which enables our systematic category-wise analyses. These analyses lead to new findings that even challenge existing knowledge and also help determine the optimal attack hyperparameters for our attack-wise comprehensive evaluation. We also pay particular attention to stealthiness, by adopting diverse imperceptibility metrics and looking into new, finer-grained characteristics. Overall, our new insights into transferability and stealthiness lead to actionable good practices for future evaluations.

While advanced machine learning (ML) models are deployed in numerous real-world applications, previous works demonstrate these models have security and privacy vulnerabilities. Various empirical research has been done in this field. However, most of the experiments are performed on target ML models trained by the security researchers themselves. Due to the high computational resource requirement for training advanced models with complex architectures, researchers generally choose to train a few target models using relatively simple architectures on typical experiment datasets. We argue that to understand ML models' vulnerabilities comprehensively, experiments should be performed on a large set of models trained with various purposes (not just the purpose of evaluating ML attacks and defenses). To this end, we propose using publicly available models with weights from the Internet (public models) for evaluating attacks and defenses on ML models. We establish a database, namely SecurityNet, containing 910 annotated image classification models. We then analyze the effectiveness of several representative attacks/defenses, including model stealing attacks, membership inference attacks, and backdoor detection on these public models. Our evaluation empirically shows the performance of these attacks/defenses can vary significantly on public models compared to self-trained models. We share SecurityNet with the research community. and advocate researchers to perform experiments on public models to better demonstrate their proposed methods' effectiveness in the future.

Large Language Models (LLMs) are powerful tools for natural language processing, enabling novel applications and user experiences. However, to achieve optimal performance, LLMs often require adaptation with private data, which poses privacy and security challenges. Several techniques have been proposed to adapt LLMs with private data, such as Low-Rank Adaptation (LoRA), Soft Prompt Tuning (SPT), and In-Context Learning (ICL), but their comparative privacy and security properties have not been systematically investigated. In this work, we fill this gap by evaluating the robustness of LoRA, SPT, and ICL against three types of well-established attacks: membership inference, which exposes data leakage (privacy); backdoor, which injects malicious behavior (security); and model stealing, which can violate intellectual property (privacy and security). Our results show that there is no silver bullet for privacy and security in LLM adaptation and each technique has different strengths and weaknesses.

We focus on learning adversarially robust classifiers under a cost-sensitive scenario, where the potential harm of different classwise adversarial transformations is encoded in a binary cost matrix. Existing methods are either empirical that cannot certify robustness or suffer from inherent scalability issues. In this work, we study whether randomized smoothing, a more scalable robustness certification framework, can be leveraged to certify cost-sensitive robustness. Built upon a notion of cost-sensitive certified radius, we show how to adapt the standard randomized smoothing certification pipeline to produce tight robustness guarantees for any cost matrix. In addition, with fine-grained certified radius optimization schemes specifically designed for different data subgroups, we propose an algorithm to train smoothed classifiers that are optimized for cost-sensitive robustness. Extensive experiments on image benchmarks and a real-world medical dataset demonstrate the superiority of our method in achieving significantly improved performance of certified cost-sensitive robustness while having a negligible impact on overall accuracy.

Large language models (LLMs) have demonstrated superior performance compared to previous methods on various tasks, and often serve as the foundation models for many researches and services. However, the untrustworthy third-party LLMs may covertly introduce vulnerabilities for downstream tasks. In this paper, we explore the vulnerability of LLMs through the lens of backdoor attacks. Different from existing backdoor attacks against LLMs, ours scatters multiple trigger keys in different prompt components. Such a Composite Backdoor Attack (CBA) is shown to be stealthier than implanting the same multiple trigger keys in only a single component. CBA ensures that the backdoor is activated only when all trigger keys appear. Our experiments demonstrate that CBA is effective in both natural language processing (NLP) and multimodal tasks. For instance, with $3\%$ poisoning samples against the LLaMA-7B model on the Emotion dataset, our attack achieves a $100\%$ Attack Success Rate (ASR) with a False Triggered Rate (FTR) below $2.06\%$ and negligible model accuracy degradation. The unique characteristics of our CBA can be tailored for various practical scenarios, e.g., targeting specific user groups. Our work highlights the necessity of increased security research on the trustworthiness of foundation LLMs.

Fine-tuning large pre-trained computer vision models is infeasible for resource-limited users. Visual prompt learning (VPL) has thus emerged to provide an efficient and flexible alternative to model fine-tuning through Visual Prompt as a Service (VPPTaaS). Specifically, the VPPTaaS provider optimizes a visual prompt given downstream data, and downstream users can use this prompt together with the large pre-trained model for prediction. However, this new learning paradigm may also pose security risks when the VPPTaaS provider instead provides a malicious visual prompt. In this paper, we take the first step to explore such risks through the lens of backdoor attacks. Specifically, we propose BadVisualPrompt, a simple yet effective backdoor attack against VPL. For example, poisoning $5\%$ CIFAR10 training data leads to above $99\%$ attack success rates with only negligible model accuracy drop by $1.5\%$. In particular, we identify and then address a new technical challenge related to interactions between the backdoor trigger and visual prompt, which does not exist in conventional, model-level backdoors. Moreover, we provide in-depth analyses of seven backdoor defenses from model, prompt, and input levels. Overall, all these defenses are either ineffective or impractical to mitigate our BadVisualPrompt, implying the critical vulnerability of VPL.

We consider availability data poisoning attacks, where an adversary aims to degrade the overall test accuracy of a machine learning model by crafting small perturbations to its training data. Existing poisoning strategies can achieve the attack goal but assume the victim to employ the same learning method as what the adversary uses to mount the attack. In this paper, we argue that this assumption is strong, since the victim may choose any learning algorithm to train the model as long as it can achieve some targeted performance on clean data. Empirically, we observe a large decrease in the effectiveness of prior poisoning attacks if the victim uses a different learning paradigm to train the model and show marked differences in frequency-level characteristics between perturbations generated with respect to different learners and attack methods. To enhance the attack transferability, we propose Transferable Poisoning, which generates high-frequency poisoning perturbations by alternately leveraging the gradient information with two specific algorithms selected from supervised and unsupervised contrastive learning paradigms. Through extensive experiments on benchmark image datasets, we show that our transferable poisoning attack can produce poisoned samples with significantly improved transferability, not only applicable to the two learners used to devise the attack but also for learning algorithms and even paradigms beyond.

Recent studies have shown that deep neural networks are vulnerable to adversarial examples. Numerous defenses have been proposed to improve model robustness, among which adversarial training is most successful. In this work, we revisit the robust overfitting phenomenon. In particular, we argue that overconfident models produced during adversarial training could be a potential cause, supported by the empirical observation that the predicted labels of adversarial examples generated by models with better robust generalization ability tend to have significantly more even distributions. Based on the proposed definition of adversarial certainty, we incorporate an extragradient step in the adversarial training framework to search for models that can generate adversarially perturbed inputs with lower certainty, further improving robust generalization. Our approach is general and can be easily combined with other variants of adversarial training methods. Extensive experiments on image benchmarks demonstrate that our method effectively alleviates robust overfitting and is able to produce models with consistently improved robustness.

The way users acquire information is undergoing a paradigm shift with the advent of ChatGPT. Unlike conventional search engines, ChatGPT retrieves knowledge from the model itself and generates answers for users. ChatGPT's impressive question-answering (QA) capability has attracted more than 100 million users within a short period of time but has also raised concerns regarding its reliability. In this paper, we perform the first large-scale measurement of ChatGPT's reliability in the generic QA scenario with a carefully curated set of 5,695 questions across ten datasets and eight domains. We find that ChatGPT's reliability varies across different domains, especially underperforming in law and science questions. We also demonstrate that system roles, originally designed by OpenAI to allow users to steer ChatGPT's behavior, can impact ChatGPT's reliability in an imperceptible way. We further show that ChatGPT is vulnerable to adversarial examples, and even a single character change can negatively affect its reliability in certain cases. We believe that our study provides valuable insights into ChatGPT's reliability and underscores the need for strengthening the reliability and security of large language models (LLMs).

State-of-the-art Text-to-Image models like Stable Diffusion and DALLE$\cdot$2 are revolutionizing how people generate visual content. At the same time, society has serious concerns about how adversaries can exploit such models to generate unsafe images. In this work, we focus on demystifying the generation of unsafe images and hateful memes from Text-to-Image models. We first construct a typology of unsafe images consisting of five categories (sexually explicit, violent, disturbing, hateful, and political). Then, we assess the proportion of unsafe images generated by four advanced Text-to-Image models using four prompt datasets. We find that these models can generate a substantial percentage of unsafe images; across four models and four prompt datasets, 14.56% of all generated images are unsafe. When comparing the four models, we find different risk levels, with Stable Diffusion being the most prone to generating unsafe content (18.92% of all generated images are unsafe). Given Stable Diffusion's tendency to generate more unsafe content, we evaluate its potential to generate hateful meme variants if exploited by an adversary to attack a specific individual or community. We employ three image editing methods, DreamBooth, Textual Inversion, and SDEdit, which are supported by Stable Diffusion. Our evaluation result shows that 24% of the generated images using DreamBooth are hateful meme variants that present the features of the original hateful meme and the target individual/community; these generated images are comparable to hateful meme variants collected from the real world. Overall, our results demonstrate that the danger of large-scale generation of unsafe images is imminent. We discuss several mitigating measures, such as curating training data, regulating prompts, and implementing safety filters, and encourage better safeguard tools to be developed to prevent unsafe generation.