Publications

Automated Exposure Notication (AEN) was implemented in 2020 to supplement traditional contact tracing for COVID-19 by estimating "too close for too long" proximities of people using the service. AEN uses Bluetooth messages to privately label and recall proximity events, so that persons who were likely exposed to SARS-CoV-2 can take the appropriate steps recommended by their health care authority. This paper describes an agent-based model that estimates the effects of AEN deployment on COVID-19 caseloads and public health workloads in the context of other critical public health measures available during the COVID-19 pandemic. We selected simulation variables pertinent to AEN deployment options, varied them in accord with the system dynamics available in 2020-2021, and calculated the outcomes of key metrics across repeated runs of the stochastic multi-week simulation. SimAEN's parameters were set to ranges of observed values in consultation with public health professionals and the rapidly accumulating literature on COVID-19 transmission; the model was validated against available population-level disease metrics. Estimates from SimAEN can help public health officials determine what AEN deployment decisions (e.g., configuration, workflow integration, and targeted adoption levels) can be most effective in their jurisdiction, in combination with other COVID-19 interventions (e.g., mask use, vaccination, quarantine and isolation periods).

The Director's Office at MIT Lincoln Laboratory (MIT LL) requested a comprehensive study on artificial intelligence (AI) focusing on present applications and future science and technology (S&T) opportunities in the Cyber Security and Information Sciences Division (Division 5). This report elaborates on the main results from the study. Since the AI field is evolving so rapidly, the study scope was to look at the recent past and ongoing developments to lead to a set of findings and recommendations. It was important to begin with a short AI history and a lay-of-the-land on representative developments across the Department of Defense (DoD), intelligence communities (IC), and Homeland Security. These areas are addressed in more detail within the report. A main deliverable from the study was to formulate an end-to-end AI canonical architecture that was suitable for a range of applications. The AI canonical architecture, formulated in the study, serves as the guiding framework for all the sections in this report. Even though the study primarily focused on cyber security and information sciences, the enabling technologies are broadly applicable to many other areas. Therefore, we dedicate a full section on enabling technologies in Section 3. The discussion on enabling technologies helps the reader clarify the distinction among AI, machine learning algorithms, and specific techniques to make an end-to-end AI system viable. In order to understand what is the lay-of-the-land in AI, study participants performed a fairly wide reach within MIT LL and external to the Laboratory (government, commercial companies, defense industrial base, peers, academia, and AI centers). In addition to the study participants (shown in the next section under acknowledgements), we also assembled an internal review team (IRT). The IRT was extremely helpful in providing feedback and in helping with the formulation of the study briefings, as we transitioned from datagathering mode to the study synthesis. The format followed throughout the study was to highlight relevant content that substantiates the study findings, and identify a set of recommendations. An important finding is the significant AI investment by the so-called "big 6" commercial companies. These major commercial companies are Google, Amazon, Facebook, Microsoft, Apple, and IBM. They dominate in the AI ecosystem research and development (R&D) investments within the U.S. According to a recent McKinsey Global Institute report, cumulative R&D investment in AI amounts to about $30 billion per year. This amount is substantially higher than the R&D investment within the DoD, IC, and Homeland Security. Therefore, the DoD will need to be very strategic about investing where needed, while at the same time leveraging the technologies already developed and available from a wide range of commercial applications. As we will discuss in Section 1 as part of the AI history, MIT LL has been instrumental in developing advanced AI capabilities. For example, MIT LL has a long history in the development of human language technologies (HLT) by successfully applying machine learning algorithms to difficult problems in speech recognition, machine translation, and speech understanding. Section 4 elaborates on prior applications of these technologies, as well as newer applications in the context of multi-modalities (e.g., speech, text, images, and video). An end-to-end AI system is very well suited to enhancing the capabilities of human language analysis. Section 5 discusses AI's nascent role in cyber security. There have been cases where AI has already provided important benefits. However, much more research is needed in both the application of AI to cyber security and the associated vulnerability to the so-called adversarial AI. Adversarial AI is an area very critical to the DoD, IC, and Homeland Security, where malicious adversaries can disrupt AI systems and make them untrusted in operational environments. This report concludes with specific recommendations by formulating the way forward for Division 5 and a discussion of S&T challenges and opportunities. The S&T challenges and opportunities are centered on the key elements of the AI canonical architecture to strengthen the AI capabilities across the DoD, IC, and Homeland Security in support of national security.

In cyber security, achieving the desired balance between system security and system performance in dynamic threat environments is a long-standing open challenge for cyber defenders. Typically an increase in system security comes at the price of decreased system performance, and vice versa, easily resulting in systems that are misaligned to operator specified requirements for system security and performance as the threat environment evolves. We develop an online, reinforcement learning based methodology to automatically discover and maintain desired operating postures in security-performance space even as the threat environment changes. We demonstrate the utility of our approach and discover parameters enabling an agile response to a dynamic adversary in a simulated security game involving prototype cyber moving target defenses.

Address Space Layout Randomization (ASLR) can increase the cost of exploiting memory corruption vulnerabilities. One major weakness of ASLR is that it assumes the secrecy of memory addresses and is thus ineffective in the face of memory disclosure vulnerabilities. Even fine-grained variants of ASLR are shown to be ineffective against memory disclosures. In this paper we present an approach that synchronizes randomization with potential runtime disclosure. By applying rerandomization to the memory layout of a process every time it generates an output, our approach renders disclosures stale by the time they can be used by attackers to hijack control flow. We have developed a fully functioning prototype for x86_64 C programs by extending the Linux kernel, GCC, and the libc dynamic linker. The prototype operates on C source code and recompiles programs with a set of augmented information required to track pointer locations and support runtime rerandomization. Using this augmented information we dynamically relocate code segments and update code pointer values during runtime. Our evaluation on the SPEC CPU2006 benchmark, along with other applications, show that our technique incurs a very low performance overhead (2.1% on average).

Computer networks are present throughout all sectors of our critical infrastructure and these networks are under a constant threat of cyber attack. One prevalent computer network threat takes advantage of unauthorized, and thus insecure, hardware on a network. This paper presents a prototype simulation system for network risk assessment that is intended for use by administrators to simulate and evaluate varying network environments and attacker/defender scenarios with respect to authorized and unauthorized hardware. The system is built on the agent-based modeling paradigm and captures emergent system dynamics that result from the interactions of multiple network agents including regular and administrator users, attackers, and defenders in a network environment. The agent-based system produces both metrics and visualizations that provide insights into network security risk and serve to guide the search for efficient policies and controls to protect a network from attacks related to unauthorized hardware. The simulation model is unique in the current literature both for its network threat model and its visualized agent-based approach. We demonstrate the model via a case study that evaluates risk for several candidate security policies on a representative computer network.

Moving Target (MT) defenses have been proposed as a gamechanging approach to rebalance the security landscape in favor of the defender. MT techniques make systems less deterministic, less static, and less homogeneous in order to increase the level of effort required to achieve a successful compromise. However, a number of challenges in achieving effective movement lead to weaknesses in MT techniques that can often be used by the attackers to bypass or otherwise nullify the impact of that movement. In this paper, we propose that these challenges can be grouped into three main types: coverage, unpredictability, and timeliness. We provide a description of these challenges and study how they impact prominent MT techniques. We also discuss a number of other considerations faced when designing and deploying MT defenses.

A model of strategy formulation is used to study how an adaptive attacker learns to overcome a moving target cyber defense. The attacker-defender interaction is modeled as a game in which a defender deploys a temporal platform migration defense. Against this defense, a population of attackers develop strategies specifying the temporal ordering of resource investments that bring targeted zero-day exploits into existence. Attacker response to two defender temporal platform migration scheduling policies are examined. In the first defender scheduling policy, the defender selects the active platform in each match uniformly at random from a pool of available platforms. In the second policy the defender schedules each successive platform to maximize the diversity of the source code presented to the attacker. Adaptive attacker response strategies are modeled by finite state machine (FSM) constructs that evolve during simulated play against defender strategies via an evolutionary algorithm. It is demonstrated that the attacker learns to invest heavily in exploit creation for the platform with the least similarity to other platforms when faced with a diversity defense, while avoiding investment in exploits for this least similar platform when facing a randomization defense. Additionally, it is demonstrated that the diversity-maximizing defense is superior for shorter duration attacker-defender engagements, but performs sub-optimally in extended attacker-defender interactions.

Moving-target (MT) techniques seek to randomize system components to reduce the likelihood of a successful attack, add dynamics to a system to reduce the lifetime of an attack, and diversify otherwise homogeneous collections of systems to limit the damage of a large-scale attack. In this article, we review the five dominant domains of MT techniques, consider the advantages and weaknesses of each, and make recommendations for future research.

Since the introduction of return-oriented programming, increasingly compiles defenses and subtle attacks that bypass them have been proposed. Unfortunately the lack of a unifying threat model among code reuse security papers makes it difficult to evaluate the effectiveness of defenses, and answer critical questions about the interoperability, composability, and efficacy of existing defensive techniques. For example, what combination of defenses protect against every known avenue of code reuse? What is the smallest set of such defenses? In this work, we study the space of code reuse attacks by building a formal model of attacks and their requirements, and defenses and their assumptions. We use a SAT solver to perform scenario analysis on our model in two ways. First, we analyze the defense configurations of a real-world system. Second, we reason about hypothetical defense bypasses. We prove by construction that attack extensions implementing the hypothesized functionality are possible even if a 'perfect' version of the defense is implemented. Our approach can be used to formalize the process of threat model definition, analyze defense configurations, reason about composability and efficacy, and hypothesize about new attacks and defenses.

In this paper, we present a method for detecting malicious activity within networks of interest. We leverage prior community detection work by propagating threat probabilities across graph nodes, given an initial set of known malicious nodes. We enhance prior work by employing constraints which remove the adverse effect of cyclic propagation that is a byproduct of current methods. We demonstrate the effectiveness of Probabilistic Threat Propagation on the task of detecting malicious web destinations.