A Deep Learning-based Framework for Conducting Stealthy Attacks in Industrial Control Systems

Feng, Chengjian; Li, Tingting; Zhu, Zhanxing; Chana, Deeph

doi:10.48550/arxiv.1709.06397

Cited by 12 publications

(18 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Generative models can be used to learn the distribution of normal traffic, and then automate the generation of attacks that look benign. For example, Feng et al [75] use deep learning to automatically learn what normal traffic entering the system looks like, and then generate samples from this distribution. Lin et al [142] use GAN's training method to train a generator against a discriminator, creating adversarial traffic samples that can evade the target detector.…”

Section: Intrusion Attacksmentioning

confidence: 99%

Generative Models for Security: Attacks, Defenses, and Opportunities

Bauer¹,

Bindschaedler²

2021

Preprint

View full text Add to dashboard Cite

Generative models learn the distribution of data from a sample dataset and can then generate new data instances. Recent advances in deep learning has brought forth improvements in generative model architectures, and some state-of-the-art models can (in some cases) produce outputs realistic enough to fool humans.We survey recent research at the intersection of security and privacy and generative models. In particular, we discuss the use of generative models in adversarial machine learning, in helping automate or enhance existing attacks, and as building blocks for defenses in contexts such as intrusion detection, biometrics spoofing, and malware obfuscation. We also describe the use of generative models in diverse applications such as fairness in machine learning, privacy-preserving data synthesis, and steganography. Finally, we discuss new threats due to generative models: the creation of synthetic media such as deepfakes that can be used for disinformation. INTRODUCTIONGenerative models learn to characterize the distribution of data using only samples from it and then generate new data instances from this distribution. Although generative models are not new, the deep learning revolution has reinvigorated research into generative model architectures, and deep generative models using state-of-the-art architectures can now produce output that is sometimes indistinguishable from real-world data. Along with this, comes a host of new issues and opportunities relating to security and privacy.This paper provides a comprehensive survey of research at the intersection of generative models and security and privacy. In particular, we describe the recent use of generative models in adversarial machine learning. We also discuss applications such as producing adversarial examples without perturbations, steganography, and privacy-preserving data synthesis. We show how a number of attacks and defenses for various cybersecurity problems such as password generation, intrusion, and malware detection, can benefit from generative models because of their ability to learn the distribution of the training data. By characterizing the data distribution, generative models help practitioners better understand phenomena they are studying and supplement existing attack and defense techniques. Finally, we discuss the extent to which deep generative models present new threats when used to produce synthetic media.The increased intensity of work at the intersection of generative models and security/privacy is evidenced by a growing body of literature. This is illustrated in Fig. 1, which shows the number of papers (published and pre-prints) on this topic from 2000 to 2020.This survey is structured as follows. Section 2 provides a brief overview of generative model architectures, as well as a discussion of metrics used to evaluate generative models. Section 3, we survey the use of generative models for intrusion detection, malware detection, and biometric spoofing. In Section 4, we describe how generative models can be used to create synthetic dat...

show abstract

Section: Intrusion Attacksmentioning

confidence: 99%

Generative Models for Security: Attacks, Defenses, and Opportunities

Bauer¹,

Bindschaedler²

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…These schemes make use of publicly available attack dataset [11] and modify those attacks to generate a limited number of stealthy attacks for the specific detection method. In [10,15], the authors used Generative Adversarial Networks (GANs) for learning anomaly detector classifiers and for the generation of malicious sensor measurements that will go undetected. A gradient-based adversarial attack scheme was proposed in [13] which was able to deceive Recurrent Neural Network (RNN) based anomaly detector in two-real world CPS namely WADI [4] and SWaT [17].…”

Section: Related Workmentioning

confidence: 99%

Attack Rules: An Adversarial Approach to Generate Attacks for Industrial Control Systems using Machine Learning

Umer,

Ahmed,

Jilani

et al. 2021

Preprint

View full text Add to dashboard Cite

Adversarial learning is used to test the robustness of machine learning algorithms under attack and create attacks that deceive the anomaly detection methods in Industrial Control System (ICS). Given that security assessment of an ICS demands that an exhaustive set of possible attack patterns is studied, in this work, we propose an association rule mining-based attack generation technique. The technique has been implemented using data from a Secure Water Treatment plant. The proposed technique was able to generate more than 300,000 attack patterns constituting a vast majority of new attack vectors which were not seen before. Automatically generated attacks improve our understanding of the potential attacks and enable the design of robust attack detection techniques.

show abstract

“…A number of recent studies have focused on evasion attacks on CPS anomaly detectors. In [22], the authors showed that generative adversarial networks (GANs) can be used for realtime learning of an unknown ICS anomaly detector (more specifically, a classifier) and for the generation of malicious sensor measurements that will go undetected.…”

Section: Related Workmentioning

confidence: 99%

“…First and foremost, all of the abovementioned papers examined evasion attacks, while our research focuses on poisoning attacks. Second, [22], [6], [8] considered a threat model in which the attacker manipulates the detector's input data in addition to manipulating the sensor data fed to the PLC. Such a model provides a lot of freedom for the adversary to make changes to both types of data.…”

Section: Related Workmentioning

confidence: 99%

Can't Boil This Frog: Robustness of Online-Trained Autoencoder-Based Anomaly Detectors to Adversarial Poisoning Attacks

Kravchik,

Shabtai

2020

Preprint

View full text Add to dashboard Cite

In recent years, a variety of effective neural network-based methods for anomaly and cyber attack detection in industrial control systems (ICSs) have been demonstrated in the literature. Given their successful implementation and widespread use, there is a need to study adversarial attacks on such detection methods to better protect the systems that depend upon them. The extensive research performed on adversarial attacks on image and malware classification has little relevance to the physical system state prediction domain, which most of the ICS attack detection systems belong to. Moreover, such detection systems are typically retrained using new data collected from the monitored system, thus the threat of adversarial data poisoning is significant, however this threat has not yet been addressed by the research community. In this paper, we present the first study focused on poisoning attacks on online-trained autoencoder-based attack detectors. We propose two algorithms for generating poison samples, an interpolation-based algorithm and a backgradient optimization-based algorithm, which we evaluate on both synthetic and real-world ICS data. We demonstrate that the proposed algorithms can generate poison samples that cause the target attack to go undetected by the autoencoder detector, however the ability to poison the detector is limited to a small set of attack types and magnitudes. When the poison-generating algorithms are applied to the popular SWaT dataset, we show that the autoencoder detector trained on the physical system state data is resilient to poisoning in the face of all ten of the relevant attacks in the dataset. This finding suggests that neural networkbased attack detectors used in the cyber-physical domain are more robust to poisoning than in other problem domains, such as malware detection and image processing.

show abstract

A Deep Learning-based Framework for Conducting Stealthy Attacks in Industrial Control Systems

Cited by 12 publications

References 28 publications

Generative Models for Security: Attacks, Defenses, and Opportunities

Generative Models for Security: Attacks, Defenses, and Opportunities

Attack Rules: An Adversarial Approach to Generate Attacks for Industrial Control Systems using Machine Learning

Can't Boil This Frog: Robustness of Online-Trained Autoencoder-Based Anomaly Detectors to Adversarial Poisoning Attacks

Contact Info

Product

Resources

About