Search

Identifying an attacker in as much detail as possible (e.g., what their goals, capabilities, and tactics are) can lead to better defensive strategies for the defender and may eventually help with the attribution of attacks. On the other hand, attackers try to avoid both detection and identification, blending in or appearing to be a different type of attacker. However, the algorithm does not scale very well for large network instances. Next, we present a scalable version of the algorithm by reducing the action space considering domain knowledge and other heuristics. Our initial experiments show that the scalable version performs reasonably well compared to the nonscalable version.

Adversarial evasion attacks challenge the integrity of machine learning models by creating out-of-distribution samples that are consistently misclassified by these models. While a variety of detection and mitigation approaches have been proposed, they are typically defeated by designing even more sophisticated attacks. One of the most promising group of such approaches is based on creating multiple diversified machine learning models and leveraging their ensemble properties for detection and mitigation of adversarial attacks in a dynamic "moving target'' fashion. However, an efficient implementation of such approaches imposes heavy computation cost for designing and enforcing diversity of multiple classifiers and then training a significant number of them. This paper proposes a scalable modification of dynamic ensemble approach that provides (i) a combinatorial boosting of the number of diversified classifiers which provides an exponentially expanded scope of reliable decisions for dynamic "moving target'' defense, and (ii) robust methods for fusion of ensemble decisions of the resulting classifiers and their combinations towards enhanced confidence in classification decisions in both benign and adversarial scenarios. Two versions of the approach were implemented and tested for machine learning models operating in two different modalities (network intrusion detection and color image classification). Both show significant increase of resiliency against adversarial evasion attacks with moderate to low impact on benign performance of defended machine learning model. For network modality, different versions of approach improved the benign accuracy from 98% to 100% while raising the adversarial accuracy from 0% to 90%-95%; for image modality, benign accuracy remained at the same level of 90% while the adversarial accuracy improved from 0% to about 85%.

In recent years, persistent cyber adversaries have developed increasingly sophisticated techniques to evade detection. Once adversaries have established a foothold within the target network, using seemingly-limited passive reconnaissance techniques, they can develop significant network reconnaissance capabilities. Cyber deception has been recognized as a critical capability to defend against such adversaries, but, without an accurate model of the adversary’s reconnaissance behavior, current approaches are ineffective against advanced adversaries. To address this gap, we propose a novel model to capture how advanced, stealthy adversaries acquire knowledge about the target network and establish and expand their foothold within the system. This model quantifies the cost and reward, from the adversary’s perspective, of compromising and maintaining control over target nodes. We evaluate our model through simulations in the CyberVAN testbed, and indicate how it can guide the development and deployment of future defensive capabilities, including high-interaction honeypots, so as to influence the behavior of adversaries and steer them away from critical resources.

In today’s IT landscape, organizations are increasingly exposed to an array of novel and sophisticated threats—including advanced persistent threats (APTs) and distributed denial-of-service (DDoS) attacks—which can bypass traditional defenses and persist in target systems indefinitely. Threat actors often rely on networks of compromised and remotely controlled hosts, known as botnets, to execute a number of different cyberattacks and engage in criminal or unauthorized activities. Protecting sensitive and mission-critical data from competitors, state actors, and organized crime has become increasingly critical for the well-being of many organizations. A promising approach to botnet detection and mitigation relies on moving target defense (MTD), a novel and game-changing approach to cyber defense. MTD creates asymmetric uncertainty, providing the defender with a tactical advantage over the attacker. MTD techniques are designed to continuously change or shift a system’s attack surface, thus increasing cost and complexity for the threat actors. We show how the botnet detection and mitigation problem can be decomposed in three related and relatively simpler challenges, and how these challenges can be effectively tackled adopting an MTD approach, ultimately limiting the ability of a botnet to persist within a target system.

We study the use of deception in attack graph-based Stackelberg security games. In our setting, in addition to allocating defensive resources to protect important targets from attackers, the defender can strategically manipulate the attack graph through three main types of deceptive actions. We show that finding the optimal deception and defense strategy is at least NP-hard. We provide two techniques for efficiently solving this problem: a mixed-integer linear program for layered directed acyclic graphs (DAGs) and neural architecture search for general DAGs. We empirically demonstrate that using deception on attack graphs gives the defender a significant advantage, and the algorithms we develop scale gracefully to medium-sized problems.

Today’s high-stakes adversarial interactions feature attackers who constantly breach the ever-improving security measures. Deception mitigates the defender’s loss by misleading the attacker to make suboptimal decisions. In order to formally reason about deception, we introduce the feature deception problem (FDP), a domain-independent model and present a learning and planning framework for finding the optimal deception strategy, taking into account the adversary’s preferences which are initially unknown to the defender. We make the following contributions. (1) We show that we can uniformly learn the adversary’s preferences using data from a modest number of deception strategies. (2) We propose an approximation algorithm for finding the optimal deception strategy given the learned preferences and show that the problem is NP-hard. (3) We perform extensive experiments to validate our methods and results. In addition, we provide a case study of the credit bureau network to illustrate how FDP implements deception on a real-world problem.

Cyber deception has emerged as a promising approach to increase the amount of effort required to conduct an attack campaign. Since the beginning of deception, several honey-based technologies have been built to defend individual portions of a network attack surface. Different honey-based technologies can be combined to further increase attackers’ cost and elicit behaviors from them which facilitate understanding their intentions and capabilities. Combining different deceptive elements would create a deceptive network surface. As attackers vary in their intentions and capabilities, presenting them with a one-size-fits-all deceptive network surface is inadequate. Therefore, there is clear need for dynamic deceptive network surfaces that are tailored to protect against different adversary classes. In a resource-constrained environment, enabling large-scale monitoring, data processing, deception planning and subsequently, deploying a customized deceptive network surface in real-time will be challenging if done manually. We envision that models inspired from the autonomic computing paradigm can efficiently tackle such challenges. To enable the development of such models and provide empirical evidence to validate their efficacy, in this chapter, we will present a framework that can act as a common platform to study different autonomic computing models. The framework is built on top of an existing deception platform called ACyDS. We will describe the current platform and enumerate its capabilities such as sensing the environment and generating deceptive network surfaces. We will also show how a well-known autonomic computing architecture called MAPE-K can be realized through our framework.

Cyber training, security testing, and research and development activities are vital to improve the security posture of a network. Currently, many institutions use cyber security testbeds to conduct these activities in an isolated virtual environment. One of the important requirements for such an environment is to provide organizers (or experimenters) with a library of vulnerable network scenarios and capabilities to mount attacks against them. However, the task of preparing a vulnerable network scenario in current testbed environments is costly in time and labor, requires significant support from the testbed staff. To this end, we present a toolset called VulnerVAN that creates a vulnerable network scenario to realize an attack sequence. In this paper, we discuss the design of VulnerVAn - our proof-of-concept implementation on CyberVAN - and present a new high-level attack specification language that enables users to chain attack steps into an attack sequence. For a given attack sequence and network scenario, VulnerVAN identifies all possible attack paths through the network that can realize the attack sequence, and provides instructions to configure machines on an attack path selected by the user. VulnerVAn also provides an attack blueprint that can guide a Red team or an automated attacker to execute the attack sequence. To demonstrate VulnerVAN's capability, we consider the use-case of a typical data exfiltration attack sequence conducted by APTs and study the performance of VulnerVAn in mapping the attack sequence to different networks.

A great deal of effort is devoted to detecting the presence of cyber attacks, so that defenders can respond to protect the network and mitigate the damage of the attack. Going beyond detection, identifying in as much detail as possible what specific type of attacker the defender is facing (e.g., what their goals, capabilities, and tactics are) can lead to even better defensive strategies and may be able to help with eventual attribution of attacks. However, attackers may wish to avoid both detection and identification, blending in or appearing to be a different type of attacker. We present a game-theoretic approach for optimizing defensive deception actions (e.g., honeypots) with the specific goal of identifying specific attackers as early as possible in an attack. We present case studies showing how this approach works, and initial simulation results from a general model that captures this problem.

Organizations increasingly rely on complex networked systems to maintain operational efficiency. While the widespread adoption of network-based IT solutions brings significant benefits to both commercial and government organizations, it also exposes them to an array of novel threats. Specifically, malicious actors can use networks of compromised and remotely controlled hosts, known as botnets, to execute a number of different cyber-attacks and engage in criminal or otherwise unauthorized activities. Most notably, botnets can be used to exfiltrate highly sensitive data from target networks, including military intelligence from government agencies and proprietary data from enterprise networks. What makes the problem even more complex is the recent trend towards stealthier and more resilient botnet architectures, which depart from traditional centralized architectures and enable botnets to evade detection and persist in a system for extended periods of time. A promising approach to botnet detection and mitigation relies on Adaptive Cyber Defense (ACD), a novel and game-changing approach to cyber defense. We show that detecting and mitigating stealthy botnets is a multi-faceted problem that requires addressing multiple related research challenges, and show how an ACD approach can help us address these challenges effectively.

Cyber reconnaissance is the process of gathering information about a target network for the purpose of compromising systems within that network. Network-based deception has emerged as a promising approach to disrupt attackers' reconnaissance efforts. However, limited work has been done so far on measuring the effectiveness of network-based deception. Furthermore, given that Software-Defined Networking (SDN) facilitates cyber deception by allowing network traffic to be modified and injected on-the-fly, understanding the effectiveness of employing different cyber deception strategies is critical. In this paper, we present a model to study the reconnaissance surface of a network and model the process of gathering information by attackers as interactions with a cyber defensive system that may use deception. To capture the evolution of the attackers' knowledge during reconnaissance, we design a belief system that is updated by using a Bayesian inference method. For the proposed model, we present two metrics based on KL-divergence to quantify the effectiveness of network deception. We tested the model and the two metrics by conducting experiments with a simulated attacker in an SDN-based deception system. The results of the experiments match our expectations, providing support for the model and proposed metrics.

In this paper, we discuss using defensive cyber deception to enhance the security and dependability of network systems. We argue that existing honey- * technologies require additional underlying support to implement more effective cyber deception. We enumerate the requirements for effective cyber deception and identify promising approaches and metrics for evaluation. We describe our current research that explores the use of Software-Defined Networking (SDN) technologies to deploy cyber deception in an enterprise network setting, discuss the challenges encountered and approaches considered to address these challenges, and provide directions for future work.

Adversaries are conducting attack campaigns with increasing levels of sophistication. Additionally, with the prevalence of out-of-the-box toolkits that simplify attack operations during different stages of an attack campaign, multiple new adversaries and attack groups have appeared over the past decade. Characterizing the behavior and the modus operandi of different adversaries is critical in identifying the appropriate security maneuver to detect and mitigate the impact of an ongoing attack. To this end, in this paper, we study two characteristics of an adversary: Risk-averseness and Experience level. Risk-averse adversaries are more cautious during their campaign while fledgling adversaries do not wait to develop adequate expertise and knowledge before launching attack campaigns. One manifestation of these characteristics is through the adversary's choice and usage of attack tools. To detect these characteristics, we present multi-level machine learning (ML) models that use network data generated while under attack by different attack tools and usage patterns. In particular, for risk-averseness, we considered different configurations for scanning tools and trained the models in a testbed environment. The resulting model was used to predict the cautiousness of different red teams that participated in the Cyber Shield ‘16 exercise. The predictions matched the expected behavior of the red teams. For Experience level, we considered publicly-available remote access tools and usage patterns. We developed a Markov model to simulate usage patterns of attackers with different levels of expertise and through experiments on CyberVAN, we showed that the ML model has a high accuracy.

Modern botnets can persist in networked systems for extended periods of time by operating in a stealthy manner. Despite the progress made in the area of botnet prevention, detection, and mitigation, stealthy botnets continue to pose a significant risk to enterprises. Furthermore, existing enterprise-scale solutions require significant resources to operate effectively, thus they are not practical. In order to address this important problem in a resource-constrained environment, we propose a reinforcement learning based approach to optimally and dynamically deploy a limited number of defensive mechanisms, namely honeypots and network-based detectors, within the target network. The ultimate goal of the proposed approach is to reduce the lifetime of stealthy botnets by maximizing the number of bots identified and taken down through a sequential decision-making process. We provide a proof-of-concept of the proposed approach, and study its performance in a simulated environment. The results show that the proposed approach is promising in protecting against stealthy botnets.

Distributed denial-of-service attacks are an increasing problem facing web applications, for which many defense techniques have been proposed, including several moving-target strategies. These strategies typically work by relocating targeted services over time, increasing uncertainty for the attacker, while trying not to disrupt legitimate users or incur excessive costs. Prior work has not shown, however, whether and how a rational defender would choose a moving-target method against an adaptive attacker, and under what conditions. We formulate a denial-of-service scenario as a two-player game, and solve a restricted-strategy version of the game using the methods of empirical game-theoretic analysis. Using agent-based simulation, we evaluate the performance of strategies from prior literature under a variety of attacks and environmental conditions. We find evidence for the strategic stability of various proposed strategies, such as proactive server movement, delayed attack timing, and suspected insider blocking, along with guidelines for when each is likely to be most effective.