Miles McQueen

A scalable approach to attack graph generation

Attack graphs are important tools for analyzing security vulnerabilities in enterprise networks. Previous work on attack graphs has not provided an account of the scalability of the graph generating process, and there is often a lack of logical formalism in the representation of attack graphs, which results in the attack graph being difficult to use and understand by human beings. Pioneer work by Sheyner, et al. is the first attack-graph tool based on formal logical techniques, namely model-checking. However, when applied to moderate-sized networks, Sheyners tool encountered a significant exponential explosion problem. This paper describes a new approach to represent and generate attack graphs. We propose logical attack graphs, which directly illustrate logical dependencies among attack goals and configuration information. A logical attack graph always has size polynomial to the network being analyzed. Our attack graph generation tool builds upon MulVAL, a network security analyzer based on logical programming. We demonstrate how to produce a derivation trace in the MulVAL logic-programming engine, and how to use the trace to generate a logical attack graph in quadratic time. We show experimental evidence that our logical attack graph generation algorithm is very efficient. We have generated logical attack graphs for fully connected networks of 1000 machines using a Pentium 4 CPU with 1GB of RAM.

Resilient control systems: Next generation design research

Since digital control systems were introduced to the market more than 30 years ago, the operational efficiency and stability gained through their use have fueled our migration and ultimate dependence on them for the monitoring and control of critical infrastructure. While these systems have been designed for functionality and reliability, a hostile cyber environment and uncertainties in complex networks and human interactions have placed additional parameters on the design expectations for control systems.

Smart Grid Data Integrity Attacks

Real power injections at loads and generators, and real power flows on selected lines in a transmission network are monitored and transmitted over a SCADA network to the system operator. These are used in state estimation algorithms to make dispatch, re-balance and other energy management system [EMS] decisions. Coordinated cyber attacks on power meter readings can be designed to be undetectable by any bad data detection algorithm. These unobservable attacks present a serious threat to grid operations. Of particular interest are sparse attacks that involve the compromise of a modest number of meter readings. An efficient algorithm to find all unobservable attacks [under standard DC load flow approximations] involving the compromise of exactly two power injection meters and an arbitrary number of power meters on lines is presented. This requires O(n2m) flops for a power system with n buses and m line meters. If all lines are metered, there exist canonical forms that characterize all 3, 4, and 5-sparse unobservable attacks. These can be quickly detected with O(n2) flops using standard graph algorithms. Known-secure phasor measurement units [PMUs] can be used as countermeasures against a given collection of cyber attacks. Finding the minimum number of necessary PMUs is NP-hard. It is shown that p+1 PMUs at carefully chosen buses are sufficient to neutralize a collection of p cyber attacks.

Smart grid data integrity attacks: characterizations and countermeasures π

Coordinated cyberattacks of power meter readings can be arranged to be undetectable by any bad data detection algorithm in the power system state estimation process. These unobservable attacks present a potentially serious threat to grid operations. Of particular interest are sparse attacks that involve the compromise of a modest number of meter readings. An efficient algorithm to find all unobservable attacks [under standard DC load flow approximations] involving the compromise of exactly two power injection meters and an arbitrary number of line power meters is presented. This requires O(n2m) flops for a power system with n buses and m line meters. If all lines are metered, there exist canonical forms that characterize all 3, 4, and 5-sparse unobservable attacks. These can be quickly detected in power systems using standard graph algorithms. Known-secure phasor measurement units [PMUs] can be used as countermeasures against an arbitrary collection of cyberattacks. Finding the minimum number of necessary PMUs is NP-hard. It is shown that p + 1 PMUs at carefully chosen buses are sufficient to neutralize a collection of p cyberattacks.

Time-to-Compromise Model for Cyber Risk Reduction Estimation

We propose a new model for estimating the time to compromise a system component that is visible to an attacker. The model provides an estimate of the expected value of the time-to-compromise as a function of known and visible vulnerabilities, and attacker skill level. The time-to-compromise random process model is a composite of three subprocesses associated with attacker actions aimed at the exploitation of vulnerabilities. In a case study, the model was used to aid in a risk reduction estimate between a baseline Supervisory Control and Data Acquisition (SCADA) system and the baseline system enhanced through a specific set of control system security remedial actions. For our case study, the total number of system vulnerabilities was reduced by 86% but the dominant attack path was through a component where the number of vulnerabilities was reduced by only 42% and the time-to-compromise of that component was increased by only 13% to 30% depending on attacker skill level.

Measuring the attack surfaces of two FTP daemons

Software consumers often need to choose between different software that provide the same functionality. Today, security is a quality that many consumers, especially system administrators, care about and will use in choosing one soft- ware system over another. An attack surface metric is a security metric for comparing the relative security of similar software systems [7]. The measure of a systems attack surface is an indicator of the systems security: given two systems, we compare their attack surface measurements to decide whether one is more secure than another along each of the following three dimensions: methods, channels, and data. In this paper, we use the attack surface metric to measure the attack surfaces of two open source FTP daemons: ProFTPD 1.2.10 and Wu-FTPD 2.6.2. Our measurements show that ProFTPD is more secure along the method dimension, ProFTPD is as secure as Wu-FTPD along the channel dimension, and Wu-FTPD is more secure along the data dimension. We also demonstrate how software consumers can use the attack surface metric in making a choice between the two FTP daemons.

Quantitative Cyber Risk Reduction Estimation Methodology for a Small SCADA Control System

We propose a new methodology for obtaining a quantitative measurement of the risk reduction achieved when a control system is modified with the intent to improve cyber security defense against external attackers. The proposed methodology employs a directed graph called a compromise graph, where the nodes represent stages of a potential attack and the edges represent the expected time-to-compromise for differing attacker skill levels. Time-to-compromise is modeled as a function of known vulnerabilities and attacker skill level. The methodology was used to calculate risk reduction estimates for a specific SCADA system and for a specific set of control system security remedial actions. Despite an 86% reduction in the total number of vulnerabilities, the estimated time-to-compromise was increased only by about 3 to 30% depending on target and attacker skill level.

Improving Attack Graph Visualization through Data Reduction and Attack Grouping

Various tools exist to analyze enterprise network systems and to produce attack graphs detailing how attackers might penetrate into the system. These attack graphs, however, are often complex and difficult to comprehend fully, and a human user may find it problematic to reach appropriate configuration decisions. This paper presents methodologies that can 1) automatically identify portions of an attack graph that do not help a user to understand the core security problems and so can be trimmed, and 2) automatically group similar attack steps as virtual nodes in a model of the network topology, to immediately increase the understandability of the data. We believe both methods are important steps toward improving visualization of attack graphs to make them more useful in configuration management for large enterprise networks. We implemented our methods using one of the existing attack-graph toolkits. Initial experimentation shows that the proposed approaches can 1) significantly reduce the complexity of attack graphs by trimming a large portion of the graph that is not needed for a user to understand the security problem, and 2) significantly increase the accessibility and understandability of the data presented in the attack graph by clearly showing, within a generated visualization of the network topology, the number and type of potential attacks to which each host is exposed.

Ideal based cyber security technical metrics for control systems

Much of the worlds critical infrastructure is at risk from attack through electronic networks connected to control systems. Security metrics are important because they provide the basis for management decisions that affect the protection of the infrastructure. A cyber security technical metric is the security relevant output from an explicit mathematical model that makes use of objective measurements of a technical object. A specific set of technical security metrics are proposed for use by the operators of control systems. Our proposed metrics are based on seven security ideals associated with seven corresponding abstract dimensions of security. We have defined at least one metric for each of the seven ideals. Each metric is a measure of how nearly the associated ideal has been achieved. These seven ideals provide a useful structure for further metrics development. A case study shows how the proposed metrics can be applied to an operational control system.

Empirical Estimates and Observations of 0Day Vulnerabilities

We define a 0Day vulnerability to be any vulnerability, in deployed software, that has been discovered by at least one person but has not yet been publicly announced or patched. These 0Day vulnerabilities are of particular interest when assessing the risk to a system from exploit of vulnerabilities which are not generally known to the public or, most importantly, to the owners of the system. Using the 0Day definition given above, we analyzed the 0Day lifespans of 491 vulnerabilities and conservatively estimated that in the worst year there were on average 2500 0Day vulnerabilities in existence on any given day. Then using a small but intriguing set of 15 0Day vulnerability lifespans representing the time from actual discovery to public disclosure, we made a more aggressive estimate. In this case, we estimated that in the worst year there were, on average, 4500 0Day vulnerabilities in existence on any given day.