skip to main content

Cyber-physical system security is a significant concern in the critical infrastructure. Strong interdependencies between cyber and physical components render cyber-physical systems highly susceptible to integrity attacks such as injecting malicious data and projecting fake sensor measurements. Traditional security models partition cyber-physical systems into just two domains – high and low. This absolute partitioning is not well suited to cyber-physical systems because they comprise multiple overlapping partitions. Information flow properties, which model how inputs to a system affect its outputs across security partitions, are important considerations in cyber-physical systems. Information flows support traceability analysis that helps detect vulnerabilities and anomalous sources, contributing to the implementation of mitigation measures. This chapter describes an automated model with graph-based information flow traversal for identifying information flow paths in the Automatic Dependent Surveillance-Broadcast (ADS-B) system used in civilian aviation, and subsequently partitioning the flows into security domains. The results help identify ADS-B system vulnerabilities to failures and attacks, and determine potential mitigation measures.
Award ID(s):
Publication Date:
Journal Name:
Critical Infrastructure Protection XIV
Page Range or eLocation-ID:
Sponsoring Org:
National Science Foundation
More Like this
  1. The increasing penetration of cyber systems into smart grids has resulted in these grids being more vulnerable to cyber physical attacks. The central challenge of higher order cyber-physical contingency analysis is the exponential blow-up of the attack surface due to a large number of attack vectors. This gives rise to computational challenges in devising efficient attack mitigation strategies. However, a system operator can leverage private information about the underlying network to maintain a strategic advantage over an adversary equipped with superior computational capability and situational awareness. In this work, we examine the following scenario: A malicious entity intrudes the cyber-layermore »of a power network and trips the transmission lines. The objective of the system operator is to deploy security measures in the cyber-layer to minimize the impact of such attacks. Due to budget constraints, the attacker and the system operator have limits on the maximum number of transmission lines they can attack or defend. We model this adversarial interaction as a resource-constrained attacker-defender game. The computational intractability of solving large security games is well known. However, we exploit the approximately modular behavior of an impact metric known as the disturbance value to arrive at a linear-time algorithm for computing an optimal defense strategy. We validate the efficacy of the proposed strategy against attackers of various capabilities and provide an algorithm for a real-time implementation.« less
  2. Security of machine learning is increasingly becoming a major concern due to the ubiquitous deployment of deep learning in many security-sensitive domains. Many prior studies have shown external attacks such as adversarial examples that tamper the integrity of DNNs using maliciously crafted inputs. However, the security implication of internal threats (i.e., hardware vulnerabilities) to DNN models has not yet been well understood. In this paper, we demonstrate the first hardware-based attack on quantized deep neural networks–DeepHammer–that deterministically induces bit flips in model weights to compromise DNN inference by exploiting the rowhammer vulnerability. DeepHammer performs an aggressive bit search in themore »DNN model to identify the most vulnerable weight bits that are flippable under system constraints. To trigger deterministic bit flips across multiple pages within a reasonable amount of time, we develop novel system-level techniques that enable fast deployment of victim pages, memory-efficient rowhammering and precise flipping of targeted bits. DeepHammer can deliberately degrade the inference accuracy of the victim DNN system to a level that is only as good as random guess, thus completely depleting the intelligence of targeted DNN systems. We systematically demonstrate our attacks on real systems against 11 DNN architectures with 4 datasets corresponding to different application domains. Our evaluation shows that DeepHammer is able to successfully tamper DNN inference behavior at run-time within a few minutes. We further discuss several mitigation techniques from both algorithm and system levels to protect DNNs against such attacks. Our work highlights the need to incorporate security mechanisms in future deep learning systems to enhance the robustness against hardware-based deterministic fault injections.« less
  3. Cyber-physical systems are vulnerable to a variety of cyber, physical and cyber-physical attacks. The security of cyber-physical systems can be enhanced beyond what can be achieved through firewalls and trusted components by building trust from observed and/or expected behaviors. These behaviors can be encoded as invariants. Information flows that do not satisfy the invariants are used to identify and isolate malfunctioning devices and cyber intrusions. However, the distributed architectures of cyber-physical systems often contain multiple access points that are physically and/or digitally linked. Thus, invariants may be difficult to determine and/or computationally prohibitive to check in real time. Researchers havemore »employed various methods for determining the invariants by analyzing the designs of and/or data generated by cyber-physical systems such as water treatment plants and electric power grids. This chapter compares the effectiveness of detecting attacks on a water treatment plant using design-centric invariants versus data-centric rules, the latter generated using a variety of data mining methods. The methods are compared based on the maximization of true positives and minimization of false positives.« less
  4. Access control and information flow are the two building blocks in the design of secure software. Of the two, access control seems ubiquitous, being widely used in operating systems, databases, firewalls, servers, web applications, and so on. The successes of information flow seem less obvious, and its benefits and potential underappreciated. Yet, when it comes to defending against malicious code, access control based defenses have proved susceptible to evasion, or they end up being so restrictive as to interfere with legitimate use. In this talk, I will argue that defenses based on information flow can be more discerning, as theymore »utilize not only the operations performed but also their context, e.g., whether malicious actors could be exerting control over these operation or their key arguments. I will then describe successful applications of information flow to defend against every stage of a cyber attack campaign, including: (a) exploit mitigation for a wide range of software vulnerabilities, (b) malware containment across diverse OSes, including Linux, BSD, and Windows XP through Windows 10, and (c) attack campaign reconstruction, where we achieve a five to six orders of magnitude data reduction by applying our techniques.« less
  5. Privilege separation is an effective technique to improve software security. However, past partitioning systems do not allow programmers to make quantitative tradeoffs between security and performance. In this paper, we describe our toolchain called PM. It can automatically find the optimal boundary in program partitioning. This is achieved by solving an integer-programming model that optimizes for a user-chosen metric while satisfying the remaining security and performance constraints on other metrics. We choose security metrics to reason about how well computed partitions enforce information flow control to: (1) protect the program from low-integrity inputs or (2) prevent leakage of program secrets.more »As a result, functions in the sensitive module that fall on the optimal partition boundaries automatically identify where declassification is necessary. We used PM to experiment on a set of real-world programs to protect confidentiality and integrity; results show that, with moderate user guidance, PM can find partitions that have better balance between security and performance than partitions found by a previous tool that requires manual declassification.« less