Model stealing attacks have been successfully used in many machine learning domains, but there is little understanding of how these attacks work against models that perform malware detection. Malware detection and, in general, security domains have unique conditions. In particular, there are very strong requirements for low false positive rates (FPR). Antivirus products (AVs) that use machine learning are very complex systems to steal, malware binaries continually change, and the whole environment is adversarial by nature. This study evaluates active learning model stealing attacks against publicly available stand-alone machine learning malware classifiers and also against antivirus products. The study proposes a new neural network architecture for surrogate models (dualFFNN) and a new model stealing attack that combines transfer and active learning for surrogate creation (FFNN-TL). We achieved good surrogates of the stand-alone classifiers with up to 99\% agreement with the target models, using less than 4% of the original training dataset. Good surrogates of AV systems were also trained with up to 99% agreement and less than 4,000 queries. The study uses the best surrogates to generate adversarial malware to evade the target models, both stand-alone and AVs (with and without an internet connection). Results show that surrogate models can generate adversarial malware that evades the targets but with a lower success rate than directly using the target models to generate adversarial malware. Using surrogates, however, is still a good option since using the AVs for malware generation is highly time-consuming and easily detected when the AVs are connected to the internet.
This work presents a novel protocol for fast secure inference of neural networks applied to computer vision applications. It focuses on improving the overall performance of the online execution by deploying a subset of the model weights in plaintext on the client's machine, in the fashion of SplitNNs. We evaluate our protocol on benchmark neural networks trained on the CIFAR-10 dataset using SEAL via TenSEAL and discuss runtime and security performances. Empirical security evaluation using Membership Inference and Model Extraction attacks showed that the protocol was more resilient under the same attacks than a similar approach also based on SplitNN. When compared to related work, we demonstrate improvements of 2.5x-10x for the inference time and 14x-290x in communication costs.
In a data-driven world, datasets constitute a significant economic value. Dataset owners who spend time and money to collect and curate the data are incentivized to ensure that their datasets are not used in ways that they did not authorize. When such misuse occurs, dataset owners need technical mechanisms for demonstrating their ownership of the dataset in question. Dataset watermarking provides one approach for ownership demonstration which can, in turn, deter unauthorized use. In this paper, we investigate a recently proposed data provenance method, radioactive data, to assess if it can be used to demonstrate ownership of (image) datasets used to train machine learning (ML) models. The original paper reported that radioactive data is effective in white-box settings. We show that while this is true for large datasets with many classes, it is not as effective for datasets where the number of classes is low $(\leq 30)$ or the number of samples per class is low $(\leq 500)$. We also show that, counter-intuitively, the black-box verification technique is effective for all datasets used in this paper, even when white-box verification is not. Given this observation, we show that the confidence in white-box verification can be improved by using watermarked samples directly during the verification process. We also highlight the need to assess the robustness of radioactive data if it were to be used for ownership demonstration since it is an adversarial setting unlike provenance identification. Compared to dataset watermarking, ML model watermarking has been explored more extensively in recent literature. However, most of the model watermarking techniques can be defeated via model extraction. We show that radioactive data can effectively survive model extraction attacks, which raises the possibility that it can be used for ML model ownership verification robust against model extraction.
In this paper, we propose a novel and practical mechanism which enables the service provider to verify whether a suspect model is stolen from the victim model via model extraction attacks. Our key insight is that the profile of a DNN model's decision boundary can be uniquely characterized by its Universal Adversarial Perturbations (UAPs). UAPs belong to a low-dimensional subspace and piracy models' subspaces are more consistent with victim model's subspace compared with non-piracy model. Based on this, we propose a UAP fingerprinting method for DNN models and train an encoder via contrastive learning that takes fingerprint as inputs, outputs a similarity score. Extensive studies show that our framework can detect model IP breaches with confidence > 99.99 within only 20 fingerprints of the suspect model. It has good generalizability across different model architectures and is robust against post-modifications on stolen models.
In model extraction attacks, adversaries can steal a machine learning model exposed via a public API by repeatedly querying it and adjusting their own model based on obtained predictions. To prevent model stealing, existing defenses focus on detecting malicious queries, truncating, or distorting outputs, thus necessarily introducing a tradeoff between robustness and model utility for legitimate users. Instead, we propose to impede model extraction by requiring users to complete a proof-of-work before they can read the model's predictions. This deters attackers by greatly increasing (even up to 100x) the computational effort needed to leverage query access for model extraction. Since we calibrate the effort required to complete the proof-of-work to each query, this only introduces a slight overhead for regular users (up to 2x). To achieve this, our calibration applies tools from differential privacy to measure the information revealed by a query. Our method requires no modification of the victim model and can be applied by machine learning practitioners to guard their publicly exposed models against being easily stolen.
Nowadays, due to the breakthrough in natural language generation (NLG), including machine translation, document summarization, image captioning, etc NLG models have been encapsulated in cloud APIs to serve over half a billion people worldwide and process over one hundred billion word generations per day. Thus, NLG APIs have already become essential profitable services in many commercial companies. Due to the substantial financial and intellectual investments, service providers adopt a pay-as-you-use policy to promote sustainable market growth. However, recent works have shown that cloud platforms suffer from financial losses imposed by model extraction attacks, which aim to imitate the functionality and utility of the victim services, thus violating the intellectual property (IP) of cloud APIs. This work targets at protecting IP of NLG APIs by identifying the attackers who have utilized watermarked responses from the victim NLG APIs. However, most existing watermarking techniques are not directly amenable for IP protection of NLG APIs. To bridge this gap, we first present a novel watermarking method for text generation APIs by conducting lexical modification to the original outputs. Compared with the competitive baselines, our watermark approach achieves better identifiable performance in terms of p-value, with fewer semantic losses. In addition, our watermarks are more understandable and intuitive to humans than the baselines. Finally, the empirical studies show our approach is also applicable to queries from different domains, and is effective on the attacker trained on a mixture of the corpus which includes less than 10\% watermarked samples.
Model extraction attacks have renewed interest in the classic problem of learning neural networks from queries. In this work we give the first polynomial-time algorithm for learning arbitrary one hidden layer neural networks activations provided black-box access to the network. Formally, we show that if $F$ is an arbitrary one hidden layer neural network with ReLU activations, there is an algorithm with query complexity and running time that is polynomial in all parameters that outputs a network $F'$ achieving low square loss relative to $F$ with respect to the Gaussian measure. While a number of works in the security literature have proposed and empirically demonstrated the effectiveness of certain algorithms for this problem, ours is the first with fully polynomial-time guarantees of efficiency even for worst-case networks (in particular our algorithm succeeds in the overparameterized setting).
Recent advancements of Deep Neural Networks (DNNs) have seen widespread deployment in multiple security-sensitive domains. The need of resource-intensive training and use of valuable domain-specific training data have made these models a top intellectual property (IP) for model owners. One of the major threats to the DNN privacy is model extraction attacks where adversaries attempt to steal sensitive information in DNN models. Recent studies show hardware-based side channel attacks can reveal internal knowledge about DNN models (e.g., model architectures) However, to date, existing attacks cannot extract detailed model parameters (e.g., weights/biases). In this work, for the first time, we propose an advanced model extraction attack framework DeepSteal that effectively steals DNN weights with the aid of memory side-channel attack. Our proposed DeepSteal comprises two key stages. Firstly, we develop a new weight bit information extraction method, called HammerLeak, through adopting the rowhammer based hardware fault technique as the information leakage vector. HammerLeak leverages several novel system-level techniques tailed for DNN applications to enable fast and efficient weight stealing. Secondly, we propose a novel substitute model training algorithm with Mean Clustering weight penalty, which leverages the partial leaked bit information effectively and generates a substitute prototype of the target victim model. We evaluate this substitute model extraction method on three popular image datasets (e.g., CIFAR-10/100/GTSRB) and four DNN architectures (e.g., ResNet-18/34/Wide-ResNet/VGG-11). The extracted substitute model has successfully achieved more than 90 % test accuracy on deep residual networks for the CIFAR-10 dataset. Moreover, our extracted substitute model could also generate effective adversarial input samples to fool the victim model.
Graph Neural Networks (GNNs) have achieved promising performance in various real-world applications. Building a powerful GNN model is not a trivial task, as it requires a large amount of training data, powerful computing resources, and human expertise in fine-tuning the model. Moreover, with the development of adversarial attacks, e.g., model stealing attacks, GNNs raise challenges to model authentication. To avoid copyright infringement on GNNs, verifying the ownership of the GNN models is necessary. This paper presents a watermarking framework for GNNs for both graph and node classification tasks. We 1) design two strategies to generate watermarked data for the graph classification task and one for the node classification task, 2) embed the watermark into the host model through training to obtain the watermarked GNN model, and 3) verify the ownership of the suspicious model in a black-box setting. The experiments show that our framework can verify the ownership of GNN models with a very high probability (up to $99\%$) for both tasks. Finally, we experimentally show that our watermarking approach is robust against a state-of-the-art model extraction technique and four state-of-the-art defenses against backdoor attacks.
The extraction of process models from text refers to the problem of turning the information contained in an unstructured textual process descriptions into a formal representation,i.e.,a process model. Several automated approaches have been proposed to tackle this problem, but they are highly heterogeneous in scope and underlying assumptions,i.e., differences in input, target output, and data used in their evaluation.As a result, it is currently unclear how well existing solutions are able to solve the model-extraction problem and how they compare to each other.We overcome this issue by comparing 10 state-of-the-art approaches for model extraction in a systematic manner, covering both qualitative and quantitative aspects.The qualitative evaluation compares the analysis of the primary studies on: 1 the main characteristics of each solution;2 the type of process model elements extracted from the input data;3 the experimental evaluation performed to evaluate the proposed framework.The results show a heterogeneity of techniques, elements extracted and evaluations conducted, that are often impossible to compare.To overcome this difficulty we propose a quantitative comparison of the tools proposed by the papers on the unifying task of process model entity and relation extraction so as to be able to compare them directly.The results show three distinct groups of tools in terms of performance, with no tool obtaining very good scores and also serious limitations.Moreover, the proposed evaluation pipeline can be considered a reference task on a well-defined dataset and metrics that can be used to compare new tools. The paper also presents a reflection on the results of the qualitative and quantitative evaluation on the limitations and challenges that the community needs to address in the future to produce significant advances in this area.
Model extraction attacks are a kind of attacks where an adversary obtains a machine learning model whose performance is comparable with one of the victim model through queries and their results. This paper presents a novel model extraction attack, named TEMPEST, applicable on tabular data under a practical data-free setting. Whereas model extraction is more challenging on tabular data due to normalization, TEMPEST no longer needs initial samples that previous attacks require; instead, it makes use of publicly available statistics to generate query samples. Experiments show that our attack can achieve the same level of performance as the previous attacks. Moreover, we identify that the use of mean and variance as statistics for query generation and the use of the same normalization process as the victim model can improve the performance of our attack. We also discuss a possibility whereby TEMPEST is executed in the real world through an experiment with a medical diagnosis dataset. We plan to release the source code for reproducibility and a reference to subsequent works.
Artificial Intelligence has made a significant contribution to autonomous vehicles, from object detection to path planning. However, AI models require a large amount of sensitive training data and are usually computationally intensive to build. The commercial value of such models motivates attackers to mount various attacks. Adversaries can launch model extraction attacks for monetization purposes or step-ping-stone towards other attacks like model evasion. In specific cases, it even results in destroying brand reputation, differentiation, and value proposition. In addition, IP laws and AI-related legalities are still evolving and are not uniform across countries. We discuss model extraction attacks in detail with two use-cases and a generic kill-chain that can compromise autonomous cars. It is essential to investigate strategies to manage and mitigate the risk of model theft.
We investigate whether model extraction can be used to "steal" the weights of sequential recommender systems, and the potential threats posed to victims of such attacks. This type of risk has attracted attention in image and text classification, but to our knowledge not in recommender systems. We argue that sequential recommender systems are subject to unique vulnerabilities due to the specific autoregressive regimes used to train them. Unlike many existing recommender attackers, which assume the dataset used to train the victim model is exposed to attackers, we consider a data-free setting, where training data are not accessible. Under this setting, we propose an API-based model extraction method via limited-budget synthetic data generation and knowledge distillation. We investigate state-of-the-art models for sequential recommendation and show their vulnerability under model extraction and downstream attacks. We perform attacks in two stages. (1) Model extraction: given different types of synthetic data and their labels retrieved from a black-box recommender, we extract the black-box model to a white-box model via distillation. (2) Downstream attacks: we attack the black-box model with adversarial samples generated by the white-box recommender. Experiments show the effectiveness of our data-free model extraction and downstream attacks on sequential recommenders in both profile pollution and data poisoning settings.
Machine-learning-as-a-service (MLaaS) has attracted millions of users to their splendid large-scale models. Although published as black-box APIs, the valuable models behind these services are still vulnerable to imitation attacks. Recently, a series of works have demonstrated that attackers manage to steal or extract the victim models. Nonetheless, none of the previous stolen models can outperform the original black-box APIs. In this work, we conduct unsupervised domain adaptation and multi-victim ensemble to showing that attackers could potentially surpass victims, which is beyond previous understanding of model extraction. Extensive experiments on both benchmark datasets and real-world APIs validate that the imitators can succeed in outperforming the original black-box models on transferred domains. We consider our work as a milestone in the research of imitation attack, especially on NLP APIs, as the superior performance could influence the defense or even publishing strategy of API providers.
With proliferation of DNN-based applications, the confidentiality of DNN model is an important commercial goal. Spatial accelerators, that parallelize matrix/vector operations, are utilized for enhancing energy efficiency of DNN computation. Recently, model extraction attacks on simple accelerators, either with a single processing element or running a binarized network, were demonstrated using the methodology derived from differential power analysis (DPA) attack on cryptographic devices. This paper investigates the vulnerability of realistic spatial accelerators using general, 8-bit, number representation. We investigate two systolic array architectures with weight-stationary dataflow: (1) a 3 $\times$ 1 array for a dot-product operation, and (2) a 3 $\times$ 3 array for matrix-vector multiplication. Both are implemented on the SAKURA-G FPGA board. We show that both architectures are ultimately vulnerable. A conventional DPA succeeds fully on the 1D array, requiring 20K power measurements. However, the 2D array exhibits higher security even with 460K traces. We show that this is because the 2D array intrinsically entails multiple MACs simultaneously dependent on the same input. However, we find that a novel template-based DPA with multiple profiling phases is able to fully break the 2D array with only 40K traces. Corresponding countermeasures need to be investigated for spatial DNN accelerators.
The advance of explainable artificial intelligence, which provides reasons for its predictions, is expected to accelerate the use of deep neural networks in the real world like Machine Learning as a Service (MLaaS) that returns predictions on queried data with the trained model. Deep neural networks deployed in MLaaS face the threat of model extraction attacks. A model extraction attack is an attack to violate intellectual property and privacy in which an adversary steals trained models in a cloud using only their predictions. In particular, a data-free model extraction attack has been proposed recently and is more critical. In this attack, an adversary uses a generative model instead of preparing input data. The feasibility of this attack, however, needs to be studied since it requires more queries than that with surrogate datasets. In this paper, we propose MEGEX, a data-free model extraction attack against a gradient-based explainable AI. In this method, an adversary uses the explanations to train the generative model and reduces the number of queries to steal the model. Our experiments show that our proposed method reconstructs high-accuracy models -- 0.97$\times$ and 0.98$\times$ the victim model accuracy on SVHN and CIFAR-10 datasets given 2M and 20M queries, respectively. This implies that there is a trade-off between the interpretability of models and the difficulty of stealing them.
Machine-Learning-as-a-Service providers expose machine learning (ML) models through application programming interfaces (APIs) to developers. Recent work has shown that attackers can exploit these APIs to extract good approximations of such ML models, by querying them with samples of their choosing. We propose VarDetect, a stateful monitor that tracks the distribution of queries made by users of such a service, to detect model extraction attacks. Harnessing the latent distributions learned by a modified variational autoencoder, VarDetect robustly separates three types of attacker samples from benign samples, and successfully raises an alarm for each. Further, with VarDetect deployed as an automated defense mechanism, the extracted substitute models are found to exhibit poor performance and transferability, as intended. Finally, we demonstrate that even adaptive attackers with prior knowledge of the deployment of VarDetect, are detected by it.
Model Extraction attacks exploit the target model's prediction API to create a surrogate model in order to steal or reconnoiter the functionality of the target model in the black-box setting. Several recent studies have shown that a data-limited adversary who has no or limited access to the samples from the target model's training data distribution can use synthesis or semantically similar samples to conduct model extraction attacks. In this paper, we define the hardness degree of a sample using the concept of learning difficulty. The hardness degree of a sample depends on the epoch number that the predicted label of that sample converges. We investigate the hardness degree of samples and demonstrate that the hardness degree histogram of a data-limited adversary's sample sequences is distinguishable from the hardness degree histogram of benign users' samples sequences. We propose Hardness-Oriented Detection Approach (HODA) to detect the sample sequences of model extraction attacks. The results demonstrate that HODA can detect the sample sequences of model extraction attacks with a high success rate by only monitoring 100 samples of them, and it outperforms all previous model extraction detection methods.
Artificial neural networks (ANNs) have gained significant popularity in the last decade for solving narrow AI problems in domains such as healthcare, transportation, and defense. As ANNs become more ubiquitous, it is imperative to understand their associated safety, security, and privacy vulnerabilities. Recently, it has been shown that ANNs are susceptible to a number of adversarial evasion attacks--inputs that cause the ANN to make high-confidence misclassifications despite being almost indistinguishable from the data used to train and test the network. This work explores to what degree finding these examples maybe aided by using side-channel information, specifically switching power consumption, of hardware implementations of ANNs. A black-box threat scenario is assumed, where an attacker has access to the ANN hardware's input, outputs, and topology, but the trained model parameters are unknown. Then, a surrogate model is trained to have similar functional (i.e. input-output mapping) and switching power characteristics as the oracle (black-box) model. Our results indicate that the inclusion of power consumption data increases the fidelity of the model extraction by up to 30 percent based on a mean square error comparison of the oracle and surrogate weights. However, transferability of adversarial examples from the surrogate to the oracle model was not significantly affected.
The collection and availability of big data, combined with advances in pre-trained models (e.g., BERT, XLNET, etc), have revolutionized the predictive performance of modern natural language processing tasks, ranging from text classification to text generation. This allows corporations to provide machine learning as a service (MLaaS) by encapsulating fine-tuned BERT-based models as APIs. However, BERT-based APIs have exhibited a series of security and privacy vulnerabilities. For example, prior work has exploited the security issues of the BERT-based APIs through the adversarial examples crafted by the extracted model. However, the privacy leakage problems of the BERT-based APIs through the extracted model have not been well studied. On the other hand, due to the high capacity of BERT-based APIs, the fine-tuned model is easy to be overlearned, but what kind of information can be leaked from the extracted model remains unknown. In this work, we bridge this gap by first presenting an effective model extraction attack, where the adversary can practically steal a BERT-based API (the target/victim model) by only querying a limited number of queries. We further develop an effective attribute inference attack which can infer the sensitive attribute of the training data used by the BERT-based APIs. Our extensive experiments on benchmark datasets under various realistic settings validate the potential vulnerabilities of BERT-based APIs. Moreover, we demonstrate that two promising defense methods become ineffective against our attacks, which calls for more effective defense methods.