In recent years, infections and damage caused by malware have increased at exponential rates. At the same time, machine learning (ML) techniques have shown tremendous promise in many domains, often out performing human efforts by learning from large amounts of data. Results in the open literature suggest that ML is able to provide similar results for malware detection, achieving greater than 99% classifcation accuracy [49]. However, the same detection rates when applied in deployed settings have not been achieved. Malware is distinct from many other domains in which ML has shown success in that (1) it purposefully tries to hide, leading to noisy labels and (2) often its behavior is similar to benign software only differing in intent, among other complicating factors. This report details the reasons for the diffcultly of detecting novel malware by ML methods and offers solutions to improve the detection of novel malware.
This report presents an analysis of the Emergency Core Cooling System (ECCS) for a generic Boiling Water Reactor (BWR)-4 NPP. The Electric Power Research Institute (EPRI) developed Hazards and Consequences Analysis for Digital Systems (HAZCADS) process is applied to the ECCS and its subsystems to identify unsafe control actions (UCAs) which act as possible cyber events of concern. The analysis is performed for two design basis events: Small-break Loss of Coolant Accident (SLOCA) and general transients (TRANS), such as unintended reactor trip. In previous work, HAZCADS UCAs were combined with other cyber-attack analysis to develop a risk-informed approach; however, this was for a single system. This report explores advanced systems engineering modeling approaches to model the interactions between digital assets across multiple systems which may be targeted by cyber adversaries. The complex and interdependent design of digital systems has the potential to introduce emergent cyber properties that are generally not covered by hazard analyses nor formal nuclear Probabilistic Risk Assessment (PRA). The R&D and supporting analysis presented here explores approaches to predict and manage how interdependent system properties effect risk. To show the potential impact of a successful cyber-attack to formal PRA event tree probabilities, HAZCADS analysis was also used. HAZCADS was also used to model the automatic depressurization system (ADS) automatic actuation. This analysis extended to an integrated system analysis for common-cause failure (CCF). In this aspect, the HAZCADS analysis continued by analyzing plant design details for system connectivity in support of critical plant functions. A dependency matrix was developed to depict the integrated functionality of the interconnected systems. Areas of potential CCF are indicated. Future work could include adversary attack development to show how CCF could be caused, resulting in PRA events. Across the multiple systems that comprise the ECCS, the analysis shows that the change in such probabilities was very different between systems. This indicates that some systems have a larger potential risk impact from successful cyber-attack or digital failure, which indicates a need for these systems to have a higher priority for design and defensive measures. Furthermore, we were able to establish that a risk analysis using any arbitrary threat model establishes an ordering of components with regard to cyber-risk. This ordering can be used to influence the overall system design with an eye to lowering risk, or as a way to understand real-time risk to operational systems based on a current threat landscape. Expert knowledge of both the analysis process and the system being analyzed is required to perform a HAZCADS analysis. The need for a tiered risk analysis is demonstrated by the results of this report.
Cybersecurity for industrial control systems is an important consideration that advance reactor designers will need to consider. How cyber risk is managed is the subject of on-going research and debate in the nuclear industry. This report seeks to identify potential cyber risks for advance reactors. Identified risks are divided into absorbed risk and licensee managed risk to clearly show how cyber risks for advance reactors can potentially be transferred. Absorbed risks are risks that originate external to the licensee but may unknowingly propagate into the plant. Insights include (1) the need for unification of safety, physical security, and cybersecurity risk assessment frameworks to ensure optimal coordination of risk, (2) a quantitative risk assessment methodology in conjunction with qualitative assessments may be useful in efficiently and sufficiently managing cyber risks, and (3) cyber risk management techniques should align with a risked informed regulatory framework for advance reactors.
Seven generation III+ and generation IV nuclear reactor types, based on twelve reactor concepts surveyed, are examined using functional decomposition to extract relevant operational technology (OT) architecture information. This information is compared to existing nuclear power plants (NPPs) OT architectures to highlight novel and emergent cyber risks associated with next generation NPPs. These insights can help inform operational technology architecture requirements that will be unique to a given reactor type. Next generation NPPs have streamlined OT architectures relative to the current generation II commercial NPP fleet. Overall, without compensatory measures that provide sufficient and efficient cybersecurity controls, next generation NPPs will have increased cyber risk. Verification and validation of cyber-physical testbeds and cyber risk assessment methodologies may be an important next step to reduce cyber risk in the OT architecture design and testing phase. Coordination with safety requirements can result in OT architecture design being an iterative process.
Machine learning (ML) techniques are being used to detect increasing amounts of malware and variants. Despite successful applications of ML, we hypothesize that the full potential of ML is not realized in malware analysis (MA) due to a semantic gap between the ML and MA communities-as demonstrated in the data that is used. Due in part to the available data, ML has primarily focused on detection whereas MA is also interested in identifying behaviors. We review existing open-source malware datasets used in ML and find a lack of behavioral information that could facilitate stronger impact by ML in MA. As a first step in bridging this gap, we label existing data with behavioral information using open-source MA reports-1) altering the analysis from identifying malware to identifying behaviors, 2)~aligning ML better with MA, and 3)~allowing ML models to generalize to novel malware in a zero/few-shot learning manner. We classify the behavior of a malware family not seen during training using transfer learning from a state-of-the-art model for malware family classification and achieve 57%-84% accuracy on behavioral identification but fail to outperform the baseline set by a majority class predictor. This highlights opportunities for improvement on this task related to the data representation, the need for malware specific ML techniques, and a larger training set of malware samples labeled with behaviors.
Aging plants, efficiency goals, and safety needs are driving increased digitalization in nuclear power plants (NPP). Security has always been a key design consideration for NPP architectures, but increased digitalization and the emergence of malware such as Stuxnet, CRASHOVERRIDE, and TRITON that specifically target industrial control systems have heightened concerns about the susceptibility of NPPs to cyber attacks. The cyber security community has come to realize the impossibility of guaranteeing the security of these plants with 100% certainty, so demand for including resilience in NPP architectures is increasing. Whereas cyber security design features often focus on preventing access by cyber threats and ensuring confidentiality, integrity, and availability (CIA) of control systems, cyber resilience design features complement security features by limiting damage, enabling continued operations, and facilitating a rapid recovery from the attack in the event control systems are compromised. This paper introduces the REsilience VeRification UNit (RevRun) toolset, a software platform that was prototyped to support cyber resilience analysis of NPP architectures. Researchers at Sandia National Laboratories have recently developed models of NPP control and SCADA systems using the SCEPTRE platform. SCEPTRE integrates simulation, virtual hardware, software, and actual hardware to model the operation of cyber-physical systems. RevRun can be used to extract data from SCEPTRE experiments and to process that data to produce quantitative resilience metrics of the NPP architecture modeled in SCEPTRE. This paper details how RevRun calculates these metrics in a customizable, repeatable, and automated fashion that limits the burden placed upon the analyst. This paper describes RevRun's application and use in the context of a hypothetical attack on an NPP control system. The use case specifies the control system and a series of attacks and explores the resilience of the system to the attacks. The use case further shows how to configure RevRun to run experiments, how resilience metrics are calculated, and how the resilience metrics and RevRun tool can be used to conduct the related resilience analysis.
In recognition of their mission and in response to continuously evolving cyber threats against nuclear facilities, Department of Energy - Nuclear Energy (DOE-NE) is building the Nuclear Energy Cyber security Research, Development, and Demonstration (RD&D) Program, which includes a cyber risk management thrust. This report supports the cyber risk management thrust objective which is to deliver "Standardized methodologies for credible risk-based identification, evaluation and prioritization of digital components." In a previous task, the Sandia National Laboratories (SNL) team presented evaluation criteria and a survey to review methods to determine the most suitable techniques [1] . In this task we will identify and evaluate a series of candidate methodologies. In this report, 10 distinct methodologies are evaluated. The overall goal of this effort was to identify the current range of risk analysis techniques that were currently available, and how they could be applied, with an focus on industrial control systems (ICS). Overall, most of the techniques identified did fall into accepted risk analysis practices, though they generally addressed only one step of the multi-step risk management process. A few addressed multiple steps, but generally their treatment was superficial. This study revealed that the current state of security risk analysis in digital control systems was not comprehensive and did not support a science-based evaluation. The papers surveyed did use mathematical formulation to describe the addressed problems, and tied the models to some kind of experimental or experiential evidence as support. Most of the papers, however, did not use a rigorous approach to experimentally support the proposed models, nor did they have enough evidence supporting the efficacy of the models to statistically analyze model impact. Both of these issues stem from the difficulty and expense associated with collecting experimental data in this domain.
Neural machine learning methods, such as deep neural networks (DNN), have achieved remarkable success in a number of complex data processing tasks. These methods have arguably had their strongest impact on tasks such as image and audio processing - data processing domains in which humans have long held clear advantages over conventional algorithms. In contrast to biological neural systems, which are capable of learning continuously, deep artificial networks have a limited ability for incorporating new information in an already trained network. As a result, methods for continuous learning are potentially highly impactful in enabling the application of deep networks to dynamic data sets. Here, inspired by the process of adult neurogenesis in the hippocampus, we explore the potential for adding new neurons to deep layers of artificial neural networks in order to facilitate their acquisition of novel information while preserving previously trained data representations. Our results on the MNIST handwritten digit dataset and the NIST SD 19 dataset, which includes lower and upper case letters and digits, demonstrate that neurogenesis is well suited for addressing the stability-plasticity dilemma that has long challenged adaptive machine learning algorithms.
In this paper, we will summarize a group of architectural principles that inform the development of secure control system architectures, followed by a methodology that allows designers to understand the attack surface of components and subsystems in a way that supports the integration of these surfaces into a single attack surface. We will then show how this methodology can be used to analyze the control system attack surface from a variety of threats, including knowledgeable insiders. We close the paper with an overview of how this approach can be folded into a more rigorous mathematical analysis of the system to define the system's security posture.
Moving target defense (MTD) is an emerging paradigm in which system defenses dynamically mutate in order to decrease the overall system attack surface. Though the concept is promising, implementations have not been widely adopted. The field has been actively researched for over ten years, and has only produced a small amount of extensively adopted defenses, most notably, address space layout randomization (ASLR). This is despite the fact that there currently exist a variety of moving target implementations and proofs-of-concept. We suspect that this results from the moving target controls breaking critical system dependencies from the perspectives of users and administrators, as well as making things more difficult for attackers. As a result, the impact of the controls on overall system security is not sufficient to overcome the inconvenience imposed on legitimate system users. In this paper, we analyze a successful MTD approach. We study the control's dependency graphs, showing how we use graph theoretic and network properties to predict the effectiveness of the selected control.
Moving target defense (MTD) is an emerging paradigm in which system defenses dynamically mu- tate in order to decrease the overall system attack surface. Though the initial concept is promising, implementations have not been widely adopted. The field has been actively researched for over ten years, and has only produced a small amount of extensively adopted defenses, most notably, address space layout randomization (ASLR). This is despite the fact that there currently exist a variety of moving target implementations and proofs-of-concept. We suspect that this results from the moving target controls breaking critical system dependencies from the perspectives of users and administrators, as well as making things more difficult for attackers. As a result, the impact of the controls on overall system security is not sufficient to overcome the inconvenience imposed on legitimate system users. In this paper, we analyze a successful MTD approach. We study the con- trol's dependency graphs, showing how we use graph theoretic and network properties to predict the effectiveness of the selected control. Then, with this framework in place, the dynamic nature of some Moving Target Defenses opens the possibility of modeling them with dynamic systems approaches, such as state space representations familiar from control and systems theory. We then use this approach to develop state space models for Moving Target Defenses, provide an analysis of their properties, and suggest approaches for using them.
As cyber-security is becoming more and more important in systems development, engineers have begun to recognize and understand the types of errors they can introduce through hurried coding technique and design. This overall trend is certainly moving the software industry in the right direction and can lead to developing higher quality software-centric systems. Unfortunately, we have barely begun to examine the results of poor architectural choices, nor do we have much insight into what secure and securable architectures look like. In this paper, based on the past 40 years of work identifying specific security principles, we create a taxonomy of principles that address the abstract cyber-security needs of systems. We then tie these principles to studies of insecure systems architectures to demonstrate applicability. We close the paper with a description of other cyber-security taxonomies, how they specifically differ from this presented taxonomy, and add new principles to address gaps shown in taxonomic comparisons.
On September 5th and 6th, 2012, the Dynamic Defense Workshop: From Research to Practice brought together researchers from academia, industry, and Sandia with the goals of increasing collaboration between Sandia National Laboratories and external organizations, de ning and un- derstanding dynamic, or moving target, defense concepts and directions, and gaining a greater understanding of the state of the art for dynamic defense. Through the workshop, we broadened and re ned our de nition and understanding, identi ed new approaches to inherent challenges, and de ned principles of dynamic defense. Half of the workshop was devoted to presentations of current state-of-the-art work. Presentation topics included areas such as the failure of current defenses, threats, techniques, goals of dynamic defense, theory, foundations of dynamic defense, future directions and open research questions related to dynamic defense. The remainder of the workshop was discussion, which was broken down into sessions on de ning challenges, applications to host or mobile environments, applications to enterprise network environments, exploring research and operational taxonomies, and determining how to apply scienti c rigor to and investigating the eld of dynamic defense.