Rethinking the Security Assurances of MTD: A Gap Analysis for Network Defense

Abstract

Moving Target Defense (MTD) is a paradigm that has the potential to revolutionize the approach to network security. Although a significant number of papers have been published on the topic, there are still no standards or any dominant implementations of this concept. This article identifies and attempts to bridge the gap in understanding various aspects of MTD security while also defining the research directions necessary to implement MTD techniques in real-world scenarios. It discusses the security of key design principles of MTD, considers problems regarding applying MTD to real networks, and finally addresses threat modeling in the context of MTD. By aggregating various security aspects related to MTD, some of which have not been typically discussed in the available literature on the subject, this work aims to assist in designing future MTD schemes, help navigate around various security caveats, and highlight possible research directions that have not been sufficiently explored in the existing literature.

1. Introduction

1.1. Current Threat Landscape

Today’s world has grown entirely dependent on the existence of computer networks. In a relatively short period, they have completely revolutionized almost every aspect of our lives. Unfortunately, this has also created new attack vectors for malicious parties seeking to profit, obtain classified data, or simply disrupt other systems. This threat is ever-growing, with an average cost of a breach in millions of US dollars [1]. This can at least partially be explained by the disadvantage the defenders find themselves in. Today’s networks are vast, interconnected systems comprising thousands of different nodes and services, which run multiple software stacks. Configuring and maintaining all of this hardware and software is no easy task. A notable example of this is the Microsoft hack, where oversight on a non-production test tenant account resulted in the exfiltration of emails from senior leadership, cybersecurity, and legal teams, among other damages [2]. Even with perfect preparations and readiness, defenders are generally at a disadvantage due to the mostly reactive nature of defense mechanisms—an antivirus needs to hold a database of hashes against which every file and process is compared, just as Intrusion Detection Systems (IDSs) and Intrusion Prevention Systems (IPSs) need relevant fingerprints of network traffic. Similarly, one of the easiest ways to mitigate most vulnerabilities is to stay up to date with patches provided by the software vendor. But this is only true for issues that have been previously identified. So-called zero-day attacks, which leverage a system vulnerability that is either unknown or unaddressed by the vendor, are especially dangerous. These Advanced Persistent Threats (APTs) are exceptionally difficult to detect and even more challenging to prevent. Zero-day attacks can remain hidden for months, waiting undetected until they are eventually discovered, if they are discovered at all. Due to their rarity, they command high prices on specialized exploit acquisition platforms. Well-funded attackers, often state actors, can acquire these exploits and use them for malicious purposes, enabling them to evade detection for extended periods.

Recently, the technique known as Moving Target Defense (MTD) has gained increasing recognition. It has been formally introduced by the USA Federal Networking and Information Technology Research and Development program as a result of a search for novel approaches to cybersecurity [3]. This paradigm is noted for its capacity to thwart cyber attacks, achieved through the introduction of systemic uncertainty and the disruption of the early stages of the attack. This approach effectively increases the complexity and unpredictability of the system, thereby posing significant challenges to potential cyber threats. This can be contextualized by analyzing the Intrusion Kill Chain Model proposed by Lockheed Martin [4]. It introduces seven stages of an attack: Reconnaissance, Weaponization, Delivery, Exploitation, Installation, Command and Control, and finally Action on Target. Ideally, novel MTD techniques would disrupt the attack at the reconnaissance stage.

Although MTD was formally introduced well over a decade ago, and numerous research articles have been published on this idea, no single prevalent architecture exists. As recent surveys in this field have pointed out [5,6,7], there are no reliable and agreed-upon methods for comparing two different techniques against one another. Surprisingly, there is also no consensus regarding how to measure the security of newly proposed strategies. Instead, many authors propose their own idea on how to test the effectiveness of MTD strategies, using their own metrics and frameworks, some of which are described in Section 2.3. These, however, fall short and fail to reach a broader audience. This all contributes to the overall information chaos surrounding MTD. Many existing articles on this security paradigm focus too narrowly on a single, often very specific topic, thereby overlooking the wider context, which is further elaborated on in Section 3. This is a significant gap in the existing literature on the topic. Without solid fundamentals, there is no moving forward. For any robust MTD architecture to ever be created, there must be a thorough understanding of how and why it provides security. That requires consideration not only of MTD strategies themselves but also of the complex systems in which these are meant to operate.

1.2. Our Contribution

While prior surveys [8,9,10] have identified conceptual gaps in MTD research, our work advances these discussions by introducing a structured analytical framework focused on security assurance and practical deployment. Specifically, we organize the analysis into three dimensions: (i) security of core design principles (what/when/how to move) under realistic attacker models, (ii) implementation-level risks such as metadata leakage, forward/backward secrecy, and vulnerabilities introduced by MTD code, and (iii) operational challenges across the network life cycle, including coexistence with existing protocols, provisioning, and debugging. This framework moves beyond a descriptive survey by synthesizing these dimensions into actionable research directions and baseline criteria for standardized security metrics. Our goal is to provide a foundation for evaluating and deploying MTD in real-world networks, bridging the gap between theoretical promise and practical adoption. Within this framework, this paper presents a comprehensive analysis of security considerations associated with MTD in networked environments. It addresses several critical dimensions, including the evaluation of security metrics, the examination of core design principles, the practical challenges of deploying MTD schemes, and the development of robust threat modeling approaches. Unlike prior studies that have often treated these aspects in isolation, this work adopts a holistic perspective, situating MTD strategies within the broader operational context and identifying potential vulnerabilities that may arise from the environments in which these techniques are implemented. To the best of our knowledge, this is the first study to explore MTD security with such breadth and depth. Its primary objective is to identify existing gaps that hinder the widespread adoption of the MTD paradigm in industrial networks and to propose directions for future research. Furthermore, the paper emphasizes the need for more sophisticated attacker models and comprehensive threat modeling to ensure that MTD schemes meet their intended security objectives. By doing so, it aims to serve as both a practical guide for designers of MTD mechanisms and a foundation for advancing scholarly work in this domain. The presented paper moves from “survey and taxonomy” found in the prior literature to gap analysis and actionable research directions. In particular, it emphasizes security assurance, deployment practicality, and attacker realism, areas largely overlooked or insufficiently addressed in existing surveys.

2. Background

This section provides a brief overview of the key topics related to MTD.

2.1. MTD Taxonomy

Although few ways to categorize MTD schemes have been proposed, perhaps the most common one, and clear, has been proposed in [11]: MTD schemes can generally be described by asking three questions—what, when, and how to move. These aspects describe the most important elements of any MTD technique. Each of those, however, carries its own security considerations that need to be taken into account when designing new schemes:

• What to move, otherwise called the moving parameter, can be selected from various elements of the system. These might include [7]: IP address, port, Virtual Machines (VMs), Operating Systems (OSs), or other.
• How to move specifies the process in which a new state is chosen.
• When to move defines the decision-making process responsible for picking the right time to perform a state change. Typically, in the literature, there are three types of state change: timer-based, event-based, or a combination of both techniques, also referred to as proactive, reactive, and hybrid, respectively [6].

2.2. Attack Surface

The attack surface is defined as the set of resources in a system that can be targeted for attack [12]. This is a very broad definition, and indeed, in today’s systems, the attack surface can be very vast. Typically, it is used in conjunction with the word “reconnaissance”, a process during which attackers attempt to gather knowledge about protected assets. A well-defined attack surface in a system is crucial for effectively defending it. MTD introduced a novel concept of moving, or shifting, the attack surface within the attack space, where one or more parameters of a system are ever-changing; these elements are described in Section 2.1 under “What to move”.

An example of attack vectors on a simple network is shown in Figure 1. The attack surface may include, but is not limited to, software and firmware across all nodes, misconfigured devices, unpatched devices, externally facing assets, reused passwords, inadequate segmentation, untrained or malicious employees, insufficient authentication, shadow IT, or other vulnerabilities.

2.3. Security Metrics

Multiple approaches to metrics for measuring MTD security have been proposed. Such metrics can be useful not only to assess the schemes in terms of proposed defense parameters but also to compare one with another. Survey [5] has proposed the following grouping for them (non-security-related metrics have been omitted for brevity):

• Qualitative Metrics, or descriptive metrics, consider the quality of the defense in the MTD technique:

  –

  Security of individual defenses—model security of each individual component of the system;

  –

  Security of the ensemble—looks at a system as a whole;

  –

  As the authors point out, there are very few works that take a holistic approach and consider both individual and system security.

• Quantitative Metrics are objective metrics that often take the form of mathematical functions that can be calculated based on data from the MTD technique:

  –

  Confidentiality, Integrity, Availability (CIA) Metrics—used to measure the impact on the system during an attack;

  –

  Risk Metrics—consider risk associated with deploying the MTD technique;

  –

  Attack graph/tree—represents possible attack paths in a system;

  –

  Policy Conflict Analysis shows how different MTD countermeasures can cause security policy violations.

A completely different approach has been presented in the survey in [6], where security metrics have instead been grouped by how they impact attackers and defenders, with each metric existing in two variants:

• Attack Success Probability (ASP) and Defense Success Probability describe, respectively, the chance for an attack to succeed and for a defense to protect against it.
• Attack Utility and Defense Utility measure the gain or loss for each side by applying the MTD technique.
• Mean time to compromise a system, or mean time to failure, shows how long it takes to compromise the system or indicates its up time in the presence of attacks.
• Learning by attackers/defenders measures the degree of learning by each side after each undertaken action.
• Attack surface and system security—the first describes the amount of system resources that an attacker can use, while the latter measures security properties enhanced by MTD, in terms of CIA and “degree of vulnerability,” which is the probability that a platform selected by the attacker is vulnerable.
• Attack/defense cost indicates efficiency of the MTD scheme in either causing overhead for an attacker or increasing costs for the defender in terms of network parameters like flow table size, address space overhead, increased power consumption, or others.

It should be highlighted that, to the best of our knowledge, no common framework for measuring MTD security currently exists. The notions mentioned here represent some of the ideas presented in recent works. The absence of such a standard framework is one of the main gaps preventing the creation of standardized solutions.

3. State-of-the-Art MTD

Over the years, a significant number of articles have been published discussing different aspects of the security of MTD for networks. However, as noted in previous sections, this body of literature is characterized by a high degree of fragmentation. While research is abundant, it is currently impossible to compare different techniques meaningfully. This section categorizes the existing work through the lens of three fundamental gaps that hinder the development of a unified security perspective in MTD research.

3.1. The Standardization and Comparison Gap

The primary issue in the current state of the art is that MTD strategies effectively exist in “isolated worlds,” evaluated predominantly through ad hoc methods provided by their own creators. This lack of prevailing standards or metrics has led to significant confusion in the field.

Several researchers have attempted to address this issue by proposing new evaluation frameworks; however, these efforts often remain siloed within specific technologies. For instance, in [13], the authors proposed a new set of metrics for evaluating MTD security, focusing on network topology shuffle and software diversity. Similarly, ref. [14] examined security metrics specifically for Software-Defined Networks (SDNs), while ref. [15] attempted to quantify MTD techniques using entropy parameters such as vulnerability, attack, and attenuation. In the cloud domain, ref. [16] introduced a model based on the Shuffle, Diversity, and Redundancy (SDR) division, a categorization also utilized in a series of studies [17,18,19,20] to analyze risk and attack costs. While these individual approaches are not necessarily incorrect, the lack of a universal metric prevents any objective cross-technique comparison, leaving a critical gap in the holistic evaluation of MTD efficacy.

3.2. The Threat Modeling and Comprehensive Analysis Gap

A second significant gap relates to the depth of security analysis. As pointed out in [7], many papers entirely lack threat models or provide incomplete ones at best. This indicates a widespread failure to conduct a comprehensive analysis of an MTD approach’s impact on the overall system security.

Research often focuses on the mechanism itself rather than the attacker’s counter-response. Some exceptions exist: ref. [21] discusses IP-shuffling against sophisticated, network-aware attackers, and ref. [22] proposed models to abstract attacker behavior during different phases. Furthermore, ref. [23] utilized multidimensional attack graphs to respond to complex scenarios in cloud environments, and ref. [24] discussed the impact of MTD on the attack surface. Despite these contributions, the majority of the literature, including surveys on intelligent algorithms for optimization [25], tends to focus on defense costs and performance rather than rigorous, adversarial security modeling. This absence of a standardized threat-centric approach directly necessitates the gap analysis provided in Section 4.

3.3. The Architectural and Domain-Specific Gap

Finally, the literature reveals a significant imbalance in research focus, where MTD is often treated as a service layer or a domain-specific extension rather than an integral part of network architecture.

Cloud environments have received the most attention, with a primary focus on VM migration and containerization. Studies such as [26,27] discuss scheduling and insider threats in VMs, while refs. [28,29] explore the trade-offs between availability, reliability, and security in cloud ecosystems. In the network domain, research is often tied to specific architectures like SDNs [30,31] or Content Distribution Networks [32]. Other generic frameworks, such as MTD as a service layer [33], aim to reduce the attack surface without compromising performance. However, because these strategies are closely tied to their respective deployment environments (e.g., specific VM migration or SDN switch-over mechanics), there is a lack of research on generalized MTD security principles that can be applied across heterogeneous network infrastructures.

4. Critical Analysis of MTD Structural Gaps

This chapter moves beyond a purely descriptive aggregation of existing MTD challenges to provide a rigorous deconstruction of the fundamental security considerations inherent in dynamic defense. While the existing literature successfully identifies various hurdles, it often fails to analyze why these represent systemic vulnerabilities. By synthesizing identified challenges into a critical framework, this section aims to expose the “architectural friction” between dynamic defense paradigms and static network foundations. The goal is to establish a theoretical basis for identifying not just what the problems are, but why they represent fundamental flaws in current MTD modeling.

4.1. Key MTD Design Principles: The Entropy and Complexity Paradoxes

The most fundamental decision in designing an MTD scheme is deciding “what to move”. As established in Section 2.1, parameters range from IP addresses to “exotic” elements like software stacks. However, we identify a critical problem that underpins these choices. Selecting a moving element is not merely a tactical choice but a strategic determination of the system’s available resources, most importantly, randomness.

Current research often fails to distinguish between infinite and finite parameter spaces. The IPv6 protocol, with its vast address pool, offers nearly infinite entropy, making it an ideal candidate for neutralizing brute-force reconnaissance. In contrast, “exotic” mutations, such as shifting OS versions or software stacks, operate within an inherently finite and narrow space. This leads to the state collision vulnerability: as the number of protected nodes increases, a small parameter space guarantees state repetition. This is not just a performance issue; it is a fundamental security flaw. If an attacker compromises one state, the probability that another node resides in that same state is high. Consequently, low-entropy MTD schemes do not truly “move” the target; they merely rotate it within a predictable set of vulnerabilities.

The question of “when to move” reveals a conflict between staying close to the original MTD paradigm and event-based responses. While MTD aims to be proactive, purely trigger-driven mutation becomes reactive, the very thing MTD seeks to avoid. The hybrid approach, particularly when mutation time is host-specific [34], is likely the strongest. However, we identify a risk of critical predictability.

As illustrated in Figure 2, assigning higher mutation frequencies to critical “Red Zones” inadvertently signals asset value to an observer. An intelligent attacker can utilize traffic analysis to identify which segments are mutating fastest, thereby identifying the organization’s most valuable data. This creates a side-channel where the defense mechanism itself acts as a beacon for high-value targets, allowing the attacker to focus resources on the areas the organization is trying hardest to protect.

Furthermore, network metadata [35] remains a persistent and insufficiently recognized gap. Metadata has been used to obtain information ranging from population density [36] to the detection of cryptomining [37]. Despite works on metadata-preserving algorithms [38,39,40,41,42] and traffic-analysis resistance [43], many MTD schemes suffer from wrongful assumptions about state isolation. If an attacker is intelligent and attentive, unusual protocol properties or timing patterns may emerge as side-channels that remain invariant across mutations. Without addressing this “metadata shadow,” address mutation provides only a facade of security.

This is further exacerbated by the potential lack of Forward Secrecy [44] and Backwards Secrecy [45]. If any mathematical correlation exists between MTD states, a skilled attacker can predict future movements or decipher past communications. Finally, the added complexity of MTD code, intended to thwart zero-day exploits [9], may introduce new implementation-level vulnerabilities by expanding the attack surface. Relying on “security by obscurity” [10] is a fatal error; MTD components must be factored into the attack surface, as the controller itself becomes a structural single point of failure for both security and availability [46,47].

4.2. Deployment of MTD in Networks: Operational Friction

For MTD to achieve widespread adoption, it must coexist with the entire network life cycle—a requirement currently missing from most design-focused papers. We identify several “frictional” scenarios:

• Configuration and Tuning:There is a significant lack of empirical guidelines for setting parameters like mutation intervals, leading to ad hoc, sub-optimal deployments.
• The Reconfiguration Vulnerability: Adding new nodes (including BYOD risks) or swapping hardware requires MTD-aware provisioning. This process must be adequately secured to prevent malicious parties from introducing compromised hardware into the dynamic environment.
• Retirement Risks: Decommissioning devices requires proper handling of MTD secrets to prevent compromising the network’s history.
• Protocol Coexistence: MTD fundamentally breaks the implicit trust in network persistence. Stateful protocols like TCP assume identity remains constant; mid-handshake mutations can mimic TCP SYN flooding DoS attacks [48], wasting server resources.
• QoS and Stability: MTD may exacerbate Quality of Service (QoS) [49] challenges for delay-sensitive protocols like VoIP and video streaming.

4.3. Threat Modeling: Deconstructing Attacker Assumptions

The most glaring flaw in the MTD literature is the ill-defined attacker models [7], which are often too simplistic or based on completely unrealistic assumptions. By applying Shostack’s four-step framework [50], we can expose the limitations of existing schemes.

As illustrated in Figure 3, MTD effectiveness is highly location-dependent. While it may defeat an external eavesdropper (Attacker A), it has no effect on an attacker who has gained privilege escalation on an endpoint (Attacker C). Given that phishing and stolen credentials remain primary vectors [51], this “Insider Threat Gap” is a critical shortcoming.

Furthermore, testing against active scanning (Nmap [52]) is too naive. Advanced Persistent Threats (APTs) [53,54] utilize Passive Reconnaissance [55] to remain in the shadows and learn mutation patterns over time. To move forward, research must shift toward defending against smart, adaptive attackers [6] who understand the MTD scheme and look for the mathematical logic behind the movement.

5. Discussion

The above points highlight significant gaps in assumptions regarding what MTD will face. This paradigm certainly has the potential to address some of the currently existing threats that cannot be countered in any other way, such as APTs; however, more work needs to be done in rethinking approaches to testing it in realistic ways, simulating how actual attackers might approach breaching the network. Perhaps one way to bridge that knowledge gap is to employ the services of pentesters, or so-called “ethical hackers”—a team that assumes the role of an attacker during security exercises. Given enough experience in the topic, such people can help fine-tune the architecture early on by probing it with a different mindset than the defenders. This is often a great way to find bugs before a product is released in the open, and it can help with MTD schemes as well.

Although security metrics are not directly tied to the security of any MTD scheme, they remain an important topic worth exploring, as described in Section 2.3. Properly defined evaluation methods are essential to correctly measure, rank, and identify promising MTD schemes. They might be crucial for any stakeholders who, after all, are not necessarily experts in the MTD domain, to pick a scheme that best suits the needs of their organization. Unfortunately, existing metrics are far from perfect. They are commonly very hard, if not impossible, to properly measure outside of a controlled lab environment. A perfect example of this is ASP, which may be too naive, as it can only describe effectiveness against one, often very specific, attack, but it does not consider any other possibilities. To make matters worse, it is commonly calculated using simple Nmap scans [52] as a baseline, an issue that has already been discussed in this article. Furthermore, many authors introduce their own metrics, which adds to the overall confusion. Although this problem has already been raised in existing surveys [5,6,7], very little has been done to date to remediate this problem. The following guidelines should be fulfilled by any useful security metrics:

• Possibly easy to establish, using data that can be obtained both during simulations and from the normal runtime of the network;
• Should be generic enough so that possibly many, or all, schemes can be evaluated using it;
• As a baseline, it should compare the MTD scheme to a state-of-the-art system, protected using the best available methods;
• Using common security mechanisms and terminology to evaluate the workings of the scheme, making it understandable even to people who are not experts in the MTD field.

Finally, a potentially interesting research direction for MTD is to further understand how attackers might utilize it. As is the case with any research, it can be applied for good, in this case to defend resources, but also for malicious purposes. As every medal has two sides, MTD researchers must consider how it can also be exploited by attackers. Consider a Distributed Denial of Service (DDoS) attack or a Command-and-Control (C&C) Server that is used by malware. In both cases, a malicious attacker might employ an address mutation to make blocking the source much harder or near impossible for the defenders.

6. Conclusions

MTD represents a promising paradigm for enhancing network security by introducing systemic unpredictability and complexity, which significantly complicates adversarial operations. Despite its potential, the absence of standardized frameworks and universally accepted security metrics remains a critical barrier to its widespread adoption. This paper has addressed these challenges by emphasizing three fundamental aspects. First, the design of MTD schemes requires careful consideration of core parameters—what to move, how to move it, and when to initiate state changes. Each decision introduces unique security trade-offs that influence both resilience and operational performance. Second, robust threat modeling is indispensable for evaluating the true security benefits of MTD. Effective models must account for attacker capabilities, realistic attack vectors, and adaptive strategies, moving beyond the simplistic assumptions that currently dominate the literature. Third, the development of standardized, comprehensive security metrics is urgently needed. Existing approaches are fragmented and often context-specific, which prevents meaningful comparisons across techniques and hinders the establishment of best practices. By addressing these gaps, future research can advance MTD from a conceptual framework to a practical, deployable solution for large-scale and industrial networks. This work aims to serve as a foundation for such efforts, guiding both academic inquiry and practical implementation toward a mature, standardized, and security-assured MTD ecosystem.