SEARCH · Search NASA
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
The Bayesian Attack Model (BAM) is an analytical tool designed to enhance the comprehension of adversarial activity in OT environments. BAM leverages both expert cybersecurity insights and historical data to characterize the likelihood of adversarial behavior given anomalous observable events.
This document presents a user story for the Bayesian Attack Model (BAM) tool designed to aggregate and analyze cyber-attack observables for operational technology (OT) systems. BAM aims to empower cybersecurity analysts by providing a streamlined interface for collecting observable data from various sources, enabling real-time analysis of potential adversary activity. By enhancing the response capabilities of security teams, BAM facilitates risk-informed decision-making and improves organizational security posture. This user story outlines the key functionalities, user interactions, and requirements necessary to successfully integrate BAM with other security information and event management (SIEM) technology and cybersecurity operations centers (CSOCs).
ResDesign project has developed integrated capabilities to help cyber physical system modelers and analysts to evaluate vulnerabilities and resilience of such systems using co-simulation-based attack testbed, graph-based visualization and monitoring tool, and Bayesian optimization-based co-design capability. The project demonstrates a collection of attack scenarios and use cases in an integrated software environment.
Critical infrastructure and other operational technology (OT) environments face increasing cybersecurity risks from adversarial behavior. This paper describes the development of a risk model using a Bayesian network to enhance the comprehension of observable cyber events caused by malicious activity in OT environments. The core of the Bayesian network is a process model that describes the stages of adversary behavior. The remainder of the model is based on the MITRE ATT&CK® for Industrial Control Systems (ICS) taxonomy, which includes tactics and techniques that may be used by the adversary. The observables provide evidence for adversary behavior through the intermediary technique and tactic nodes. One challenge in constructing this model is a lack of open-source data from cyber-attacks on OT systems. This paper discusses learning from limited data, the elicitation of expert opinion to construct the conditional probability tables when data is scarce, and the refinement of the most difficult conditional probabilities tables using several forms of sensitivity analyses. Finally, the Bayesian network is demonstrated using two historical case studies: the DarkSide ransomware attack on the Colonial Pipeline and the destructive cyberattack targeting the ThyssenKrupp blast furnace. Index Terms—Cybersecurity, industrial control systems, operational technology
The ability of machine learning (ML) classification models to resist small, targeted input perturbations—known as adversarial attacks—is a key measure of their safety and reliability. We show that floating-point non associativity (FPNA) coupled with asynchronous parallel programming on GPUs is sufficient to result in misclassification, without any perturbation to the input. Additionally, we show that this misclassification is particularly significant for inputs close to the decision boundary and that standard adversarial robustness results may be overestimated up to 4.6 when not considering machine-level details. We first study a linear classifier, before focusing on standard Graph Neural Network (GNN) architectures and datasets used in robustness assessments. We develop a novel black-box attack using Bayesian optimization to discover external workloads that can change the instruction scheduling which bias the output of reductions on GPUs and reliably lead to misclassification. Motivated by these results, we present a new learnable permutation (LP) gradient-based approach to learning floating-point operation orderings that lead to misclassifications. The LP approach provides a worst-case estimate in a computationally efficient manner, avoiding the need to run identical experiments tens of thousands of times over a potentially large set of possible GPU states or architectures. Finally, using instrumentation-based testing, we investigate parallel reduction ordering across different GPU architectures under external background workloads, when utilizing multi-GPU virtualization, and when applying power capping. Our results demonstrate that parallel reduction ordering varies significantly across architectures under the first two conditions, substantially increasing the search space required to fully test the effects of this parallel scheduler-based vulnerability. These results and the methods developed here can help to include machine-level considerations into adversarial robustness assessments, which can make a difference in safety and mission critical applications.