CISAF: A Framework for Estimating the Security Posture of Academic and Research Cyberinfrastructure

TL;DR

CISAF is a systematic framework designed to evaluate and improve the security posture of academic and research cyberinfrastructures by identifying risks and prioritizing mitigation strategies.

Contribution

The paper introduces CISAF, a novel, mission-centric framework tailored for assessing security risks in heterogeneous academic cyberinfrastructures.

Findings

Provides a structured approach for security analysis

Helps prioritize mitigation actions based on risk assessment

Integrates architecture, mission, and attack analysis

Abstract

Academic and research cyberinfrastructures (AR-CIs) present unique security challenges due to their collaborative nature, heterogeneous components, and the lack of practical security assessment frameworks tailored to their needs. We propose Cyber Infrastructure Security Analysis Framework (CISAF) -- a simple, systematic, mission-centric approach to analyze the security posture of a CI and prioritize mitigation actions. CISAF guides administrators through a top-down process: (1) defining unacceptable losses, (2) identifying associated system hazards and critical assets, (3) analyzing possible attack paths that target these critical assets, and (4) analyzing security mechanisms that lie on these attack paths. By combining information about the CI architecture, mission, attack vectors, and security mechanisms, CISAF provides a clear overview of potential security risks and offers valuable information to prioritize mitigation actions.

Figures

Tables

Table 1. TABLE I: Unacceptable Losses

Loss Identifier	Description
L1	Experimental data leak
L2	Experimental data loss
L3	Experimental data corruption
L4	Interruption/termination of experiments
	(node destroyed, node restarted)
L5	Misuse of infrastructure
	(e.g., crypto mining, botnet for attacking others)

Table 2. TABLE II: A summary of system hazards, their descriptions, and associated losses.

Hazard	Hazard Description	Associated Losses
H1-facility breach
H1.1-operation server subversion	The attacker can allocate or deallocate any resources, gain access to all experiments and delete or corrupt data on the experimental nodes. They can also leak that data to the outside.	L1, L2, L3, L4, L5
H1.2-network emulator subversion	The attacker becomes a person in the middle for all experimental communication. They can corrupt or drop network traffic.	L1, L2, L3
H1.3-infrapod server subversion	The attacker becomes a person in the middle for all communication between experiments and the outside, and can redirect, drop, or corrupt this traffic. They can also impersonate outside servers and gain the trust of users, leading to sensitive data loss.	L1, L2, L3
H1.4-node server subversion	The attacker can change node images, interrupt experiments, and access experiments of other users that share the same physical node server.	L1, L2, L3, L4, L5
H1.5-bastion server subversion	The attacker becomes a man-in-the-middle between the CI and the outside world. They can generate malicious traffic to the inside or to the outside, and interfere with experiments’ outside connectivity, rendering the CI unusable.	L4, L5
H2-portal breach
H2.1-storage server subversion	The attacker has access to long-term storage for all users and projects, and can delete or corrupt experimental data, or exfiltrate sensitive data. The attacker can interfere with project, group, and experiment membership to corrupt it, to delete it, or use it to gain additional privileges.	L1, L2, L3, L4, L5
H2.2-identity server subversion	The attacker has access to all user information, such as credentials and PII. They can exfiltrate it or misuse it to impersonate a user and gain access to their resources. This access can be used to corrupt or delete experimental data, interrupt experiments, or misuse resources for unauthorized activities.	L1, L2, L3, L4, L5
H2.3-certificate service subversion	The attacker can invalidate existing certificates and produce new ones, thus interfering with a user’s access to experimental nodes, or giving the attacker unauthorized access to experiments.	L1, L2, L3, L4, L5
H3-experiment breach	The attacker can interfere with the user’s experiment, exfiltrate data, delete or corrupt experimental data, and misuse resources for profit or to attack others on the Internet.	L1, L2, L3, L4, L5
H4-network-flood	An insider attacker floods the experimental or infranet network, overwhelming network resources.	L4
H5-user turned malicious	An insider attacker starts misusing the infrastructure.	L5

Table 3. TABLE III: Attack Paths of Hazard H1.3 and the Corresponding Protections

Path Number	Path	Protections
1	Node $\to$ Infrapod pod$\to$ Infrapod Servers	infrapod pod is guarded by SSH.
		Infrapod Servers are guarded by Kubernetes Pod Security Standards
		infrapod Servers’ sudo access is guarded by Linux account access standards.
2	Node $\to$ Infrapod API	Access to the admin ports is guarded by the VXLAN isolation
		Infrapod API is guarded by credentials and login authentication system

Table 4. TABLE IV: Attack Paths of H1.1 and the Corresponding Protections

Path Number	Chain	Protections
1	Node $\to$ Node Servers $\to$ Ops	Node Server protected by virtualization
		Ops guarded by SSH and Linux account access standards
		Ops regularly patched
2	Bare-metal Node $\to$ Ops	Ops guarded by SSH and Linux account access standards
		Ops regularly patched
3	Node or Bare-metal Node $\to$ Infrapod pod$\to$ Infrapod Servers $\to$ Ops	Infrapod is guarded by SSH
		Infrapod Servers are guarded by Kubernetes Pod Security
		Standards
		Infrapod Servers sudo access is guarded by credentials
		and an authentication system
		Ops guarded by SSH and Linux account access standards
4	Internet $\to$ Infrapod Servers $\to$ Ops	Infrapod Servers protected by firewall and VPN
		Ops guarded by SSH and Linux account access standards
		Ops and infrapod servers regularly patched

Table 5. TABLE V: Attack Paths of H4 and H5 and the Corresponding Protections

Path Number	Chain	Protections
1	Node $\to$ Infranet Switch flood (H4)
2	Node $\to$ Expnet Switch flood (H4)
3	Node $\to$ Cryptomining on the node (H5)
4	Node $\to$ Malicious external traffic (H5)	Firewall rules prevent outgoing traffic, except to SSH, HTTP and HTTPS ports

Full text

1 Introduction

Academic and research cyberinfrastructures (AR-CIs, for short) are complex networks, with integrated hardware, software, services, and datasets, which are used by different research communities to advance science and facilitate open access to scientific resources [nsf_cyberinfrastructure_2024]. Like any networked resource, AR-CIs are exposed to outsider and insider attacks. But unlike many commercial networks, AR-CIs face specific challenges that come from sporadic funding and an open nature of their user base.

AR-CI funding is often small and time-limited, which limits the size and expertise of their DevOps staff and the effort they can invest in security. In addition, AR-CIs are highly motivated to support various use cases, some of which may create additional security risks. Further, AR-CIs may include nonstandard, Internet-enabled devices, such as Internet of Things (IoT) devices and automated lab equipment [adams2022findings]. The complexity and security risk of an AR-CI increase with increased device diversity. Lastly, AR-CIs often run custom management software, which may not be sufficiently secure [adams2021state].

There is a great need to provide practical, custom-tailored, and easy-to-follow guidelines to AR-CIs on how to assess their security posture. Among the existing cybersecurity frameworks for IT and OT systems, NIST’s standard is one of the most widely accepted security guidelines. NIST SP 800-30 [jtf2012guide] and NIST SP 800-160 [ross2021developing] both provide cybersecurity frameworks for designing Internet-enabled systems and infrastructures. Both documents provide high-level analytic practices and processes that are applicable to IT and OT systems, but they are not immediately and directly applicable to AR-CIs. ATT&CK [mitre_attack_2025] and STRIDE [howard2006security] frameworks provide structured approaches to enumerate threats for single devices or services, but they cannot be readily extrapolated to an AR-CI’s scale and complexity. Mission Based Cyber Risk Assessments (MBCRAs) [denaray2022analysis], System-Theoretic Process Analysis for Security (STPA-Sec) [young2021intro, sahay2023comparative], and MISSION AWARE [bakirtzis2017mission] provide frameworks for mission-based security analysis for Cyber-Physical Systems (CPS), where the computational elements are tightly integrated with and control physical processes, while Cyber Analysis Visualization Environment (CAVE) [pecharich2016mission] provides one for mission-critical systems, where any failure can lead to catastrophic consequences. These security frameworks cannot be directly applied to all elements of AR-CI, and they are prohibitively time-consuming to employ, given the size of AR-CI teams and funding.

We propose Cyber Infrastructure Security Analysis Framework (CISAF) – a simple, systematic, mission-centric approach to analyze security posture of a CI and prioritize mitigation actions. CISAF makes the following contributions:

• •

Customized for Each AR-CI. AR-CIs are designed and constructed by different teams and organizations; they serve different purposes, and thus have different security challenges. CISAF starts with a high-level architecture of an AR-CI, and its mission, and customizes the rest of the analysis to these elements.

• •

Easy to use and actionable. CISAF assessment is easy for AR-CI teams to understand and apply. It produces actionable results, which can lead to tangible improvements in AR-CI security posture.

• •

Allows for Prioritization of Mitigation. CISAF analysis provides critical information to analysts to reason about different cybersecurity improvements, and select those that make the biggest reductions in their cyber risk.

CISAF complements existing analysis frameworks. It is inspired and adopts elements of MISSION CENTRIC, MBCRA, and CAVE [denaray2022analysis, bakirtzis2017mission, pecharich2016mission], specifically a mission-centric approach for prioritization of mitigation actions and attack tree construction. AR-CI teams could follow CISAF up with more in-depth analysis, using other approaches, such as ATT&CK and STRIDE, on select pieces of their infrastructure that CISAF highlighted as needing improvements.

2 Related Work

There is a moderate amount of research on the security of CI in general. The Trusted CI [jackson2021trusted_ci, adams2021state, adams2022findings] has collaborated and interacted with many CIs to help them evaluate their cybersecurity needs and develop a strategic plan to improve cybersecurity. Trusted CI proposed a cybersecurity program implementation guideline for researchers [jackson2021trusted_ci], which contains four main “pillars”–Mission Alignment, Governance, Resources, and Controls–and sixteen “Musts” for Research CI Operators. The guideline is very valuable, but it is high-level, generic, and focuses more on teaching CI researchers and operators how to reason about cybersecurity than on analyzing the security posture of an existing CI. Further, even though the guide promotes a foundational understanding of assets through documentation and classification, it does not detail a step-by-step, actionable security and threat analysis methodology that researchers and operators should follow.

Aside from their program implementation guidelines, Trusted CI also published reports that analyze the security posture of various CIs [adams2021state, adams2022findings]. In their CI software project report [adams2021state], Trusted CI interviewed CI project teams, and highlighted several bad practices for CI software projects, such as the lack of threat modeling, documentation, maintenance, vulnerability tracking, prioritization, and poor coding language choices (not type-safe, inherently insecure, or out of date). Similarly, Trusted CI’s Operational Technology report [adams2022findings] highlights the OT cybersecurity shortcomings throughout the process of digitizing and making OT technology smart. Both reports provide valuable insights into current CI security practices, but do not offer a way to prioritize solutions or tailor the analysis to a particular CI.

NIST has provided foundational guidance that shaped cybersecurity practices in various IT and OT system designs. The widely recognized Cybersecurity Framework (CSF) [nist_csf2_2024] offers a high-level structure based on core functions (Identify, Protect, Detect, Respond, Recover) that helps organizations manage and communicate cybersecurity risk. Foundational to executing such frameworks is the systematic risk assessment process [jtf2012guide]; its methodologies for analyzing threats, vulnerabilities, and impacts form a common basis for security analysis. They are frequently referenced in research and are essential for informed design decisions. Equally fundamental, particularly when developing new OT systems, are the “Secure by Design” principles promoted by [ross2021developing], guiding the integration of security throughout the system’s engineering life-cycle. Despite the undisputed benefits of establishing comprehensive security programs and guiding secure development, applying these foundational NIST standards presents practical challenges for teams assessing the security posture within AR-CIs. The high-level, organization-wide perspective of the CSF often requires significant interpretation, knowledge, research, and time to apply to specific AR-CI components or research workflows. In contrast, the approach proposed in this report is simpler and mission-focused, and thus better suited to the limited resources of AR-CI teams.

On a lower-level scale, many security frameworks provide procedural guidelines for security analysis of specific types of cyber systems, and a fine granularity. The Mission Based Cyber Risk Assessments (MBCRAs) [denaray2022analysis], such as System-Theoretic Process Analysis for Security (STPA-Sec) [young2021intro, sahay2023comparative], MISSION AWARE [bakirtzis2017mission], and CAVE [pecharich2016mission], have inspired the ideas proposed in this paper. The STPA-Sec and MISSION AWARE frameworks were originally designed for cyber-physical systems (CPS), which mainly contain sensors, control systems, communication systems, and operation monitors. These frameworks are not well-suited to the evaluation of AR-CIs. Further, STPA-Sec and MISSION AWARE are cybersecurity frameworks designed to be applied during the system design phase, following the principle of “secure by design” rather than providing continuous security estimation and support. Even though some aspects of these security design frameworks could be applied later during the operations, they are very comprehensive and would take significant time and human resources [denaray2022analysis]. CAVE is a very comprehensive and fine-grained analysis framework for critical cyber systems, such as space missions. CAVE is highly integrated and automated, pulling in data about network architecture, installed software, and security mechanisms from different existing databases. Such analysis is prohibitively costly for an AR-CI to conduct. We propose a faster, albeit much coarser-grained and manual approach, tailored specifically for AR-CIs.

Aside from the security frameworks designed for systems, there are numerous security models and knowledge bases, such as ATT&CK [mitre_attack_2025] and STRIDE [howard2006security], for individual devices’ security analysis. They can complement our CISAF approach to enumerate attacks and find mitigations for resources identified as underprotected.

3 Background

Academic and research CIs (AR-CIs) are complex and interconnected ecosystems comprising a variety of resources (software, hardware, network elements, data). Depending on their use cases, they provide different levels of access and privileges to various research communities and individuals, thereby enabling significant scientific discoveries across various domains [nsf_cyberinfrastructure_2024].

The architecture of AR-CIs varies considerably regarding user entry points, the mission of the CI, and the internal distribution of critical assets and sensitive data. Correspondingly, the degree of control granted to users differs across CI, spanning from limited actions such as file retrieval to privileged operations like superuser access to the virtual machines or interacting directly with physical systems. For example, SPHERE testbed [sphere_colab] provides VM and bare metal machines for registered users; FABRIC [fabric_knowledge_base_2023] and ACCESS [access_allocations_2025] provide VMs for registered users; and SAGE [sage_edge_apps_2025] allows uploading, building, and sharing of AI applications for deployment on Docker containers across the network of software-defined sensors with given API and schedulers. Further, ZEUS [zeus_proposal_2025] runs a yearly application cycle, which only the researchers with accepted proposal get to access the research CI; while projects like ngEHT [ngeht_opportunities_2025] and Neutron Spin Echo For the Nation [cns_neutron_2025] only provide access to limited researchers and does not have a registration portal open to the public and no publicly accessible datasets. Greater access privileges can facilitate more versatile research, but they increase the risk of a CI from cyber threats.

AR-CIs contain several unique characteristics that distinguish their security landscape from enterprise IT and OT environments. One key aspect is the difference in missions and objectives. Commercial cyber systems are operated for profit and thus have more funding, more security measures, and more detailed monitoring of their critical assets. AR-CIs are often supported by small research funds, aiming to democratize access to resources, facilitate fast research processes and/or broaden educational benefits. Often, there is a smaller technical team, and thus less stringent control and monitoring of critical resources, potentially allowing attacks and exploitations to happen more easily and without detection. AR-CI developers may also lack knowledge and resources to develop and maintain comprehensive security systems and sensors to track usage and access patterns. This can be especially challenging, because AR-CIs run custom code and services, which makes it hard to secure them using off-the-shelf software.

Another defining characteristic of AR-CIs is their strong focus on collaboration and openness inherent in research and education. Many AR-CI missions involve open sharing of research platforms, research outputs, and resources across one institution, multiple institutions or across the nation. The necessity of sharing data, computational resources, and research processes among different organizations and teams creates tensions with traditional security principles that prioritize restriction and control. Open nature of these AR-CIs opens up a larger attack surface. Balancing this open nature with security needs is a key challenge.

Thirdly, because research activities on an AR-CI often develop new ways of leveraging resources or data, this presents additional security challenges. These research activities may themselves use outdated or custom software and services, with limited support and possibly unknown vulnerabilities, requiring a highly adaptable and updated security framework to adjust for emerging security threats.

Lastly, some AR-CIs carry complex dataflows and secrets that should not be exposed to the public, such as personally identifiable information (PII) and intellectual property. Tracking, securing, and isolating these dataflows in complex interconnected systems could be a significant challenge. The complexity of these flows expands the attack surface and increases the potential for data breaches at multiple points. Robust data governance and security measures are therefore required in the security framework design.

To balance security and an AR-CI’s mission, security measures must effectively mitigate risks, without imposing overly burdensome procedures or demanding extensive security expertise from researchers whose primary focus lies elsewhere. The complex landscape discussed above highlights the critical need for understandable and actionable security posture and security risk analysis.

4 CISAF Approach

We propose a systematic, mission-centric framework to represent and reason about the security posture of an AR-CI, called Cyber Infrastructure Security Analysis Framework or CISAF. CISAF starts with: a simplified functional architecture of the given AR-CI, supplied by the CI team and a definition of the AR-CI’s mission. It proceeds to define adverse events that would jeopardize AR-CI mission, called unacceptable losses. Next, it defines hazards, attacker-driven actions on the AR-CI architecture that may cause one or multiple losses. Next, it develops an attack tree consisting of attack paths, each describing a series of steps that would let an attacker create a hazard. Finally, CISAF adds the existing security mechanisms on attack paths – this information helps AR-CI teams reason how these mechanisms disable or weaken the attack paths. Those attack paths that have a few or no security mechanisms are candidates for mitigation actions. AR-CI teams can prioritize these actions, depending on the priority of losses that reside at the end of these paths. Figure 1 illustrates CISAF’s workflow.

4.1 Simplified AR-CI Architecture

CISAF starts with a high-level, functional architecture of an AR-CI, annotated with critical assets. The architecture includes important servers and services in the AR-CI, their physical and logical connections, physical elements that are accessible by AR-CI users or outsiders, data flows that are exposed to users or outsiders, and critical data assets, such as user credentials and experimental data. Physical or logical components that have identical functionality are grouped together in the architecture.

SPHERE Testbed Architecture. We illustrate this step on the SPHERE Testbed. Figure 2 shows SPHERE’s architecture, annotated with data and service assets. There are two parts of the infrastructure - a portal, which manages user PII data and credentials, group, experiment and project access rights, and credentials that allow user access to experiments, and a facility, which allocates resources for experiments, including virtual and physical machines, and network resources. Experimental nodes can freely reach outside via an infrapod, a Kubernetes Pod on the infrapod server, and the ingress switch and users can also configure their experiments to allow outside access via the same channel. Users access experiments via the bastion server and via a user-specific, virtual gateway server. This access between the gateway server and the experiment nodes occurs via a number of intermediate switches and via the infrapod server, which creates a VLAN between a user’s gateway server and their experimental nodes.

4.2 AR-CI Mission and Losses

Mission-centric approach to cybersecurity analysis has been used in multiple frameworks, including CAVE [pecharich2016mission], STPA-Sec [sahay2023comparative], and MISSION AWARE [bakirtzis2017mission]. Given the limited AR-CI team resources, a mission-centric approach helps the team focus on assets that are critically important to the AR-CI mission.

In this step, the AR-CI team should clearly define its mission. Most AR-CIs aim to serve their research community, and can start with this statement. The statement is then expanded, based on what kind of services and resources AR-CI offers to users, and what kind of access it facilitates.

SPHERE Testbed mission. SPHERE Testbed offers access to physical and virtual machines, and custom networks between them, and it allows superuser access to these resources. The resources can reach out to the Internet, and if needed, users can open resources for access from outsiders. SPHERE’s mission is then to support these user access modes, protect user data and experimentation, and protect the Internet from harmful actions by SPHERE’s users.

Based on the mission’s definition, the AR-CI team defines and prioritizes unacceptable losses. These are events that would endanger CI’s mission, by either jeopardizing its ability to serve its users, inflicting harm to users, or lowering CI’s reputation. AR-CI team prioritizes the losses based on their projected impact on AR-CI’s mission.

**SPHERE Testbed losses.**We identified five types of losses in Table I, ordered from highest to lowest priority: experimental data leak, experimental data loss, experimental data corruption, experiment interruption, and infrastructure misuse. Data leaks are ranked the highest, because their damage cannot be mitigated – once data is stolen, it is forever open to misuse by others. Experimental data loss occurs when the user loses access to experimental data, e.g., through deletion or ransomware. Data loss is concerning if it occurs much after the experiment has concluded. It is possible that users (usually graduate students) who conducted experiments and collected the data have left the project, and this makes recovery (e.g., via new experimentation) very costly. Similarly, experimental data corruption can have a large impact on a research project, especially if it is stealthy and hard to detect, and if it occurs much after the experiment has concluded. Experiment interruption is less concerning, because a running experiment can always be restarted on SPHERE Testbed. Finally, infrastructure misuse is of concern, because it could bring reputational damage to SPHERE Testbed. It is ranked the lowest, because it does not directly jeopardize SPHERE’s mission of supporting research, while other losses do.

4.3 From Losses to Hazards

A “hazard” is a set of attacker actions on the CI that enable the attacker to create a loss. For example, leaks of experimental data (loss) can occur if an attacker gains unauthorized access to a file system holding the experimental data (hazard). As another example, experimental data corruption (loss) can occur if an attacker gains unauthorized access to an experimental node (hazard), if an attacker generates a DoS attack on the experimental network during an experiment (another hazard). This means that we may have multiple hazards that can create a given loss. The AR-CI team combines the unacceptable losses with the knowledge captured at AR-CI architecture to identify hazards of interest.

SPHERE hazards. Table II lists the hazards we identified on SPHERE for the losses in Table I.

4.4 Attack Trees And Security Mechanisms

Attack trees [pecharich2016mission] are a well-known tool to investigate the steps an attacker must take to reach some goal in a complex system. CISAF builds attack trees that show all the paths for an attacker to achieve a hazard. The starting points of the tree are the nodes over which the attackers have direct control, such as their own machines, a gateway node, or an experimental node (VM or physical). AR-CI teams should consider both insider (AR-CI users) and outsider attackers. The ending points of the attack tree would be the entities and assets that are involved in a given hazard.

Finally, for each path on the attack tree, we consider the existing security mechanisms in AR-CI that may break links on the path. If some paths have none or insufficient security mechanisms, the AR-CI team can focus their mitigation efforts on these paths. To prioritize mitigation, the AR-CI team can refer to the losses and attend first to those paths that lead to hazards associated with high-priority losses.

SPHERE Testbed attack tree and security mechanisms. We will now illustrate the attack tree and security mechanisms for SPHERE and for three hazards from Table II.

Hazard H1.3-infrapod server subversion. Table III shows the attack tree in tabular form in the second column. The first path describes attacker’s actions where they first take control of the infrapod (e.g., through leaked credentials or the exploitation of vulnerabilities in the server’s software). Afterwards, the attacker can escape the Kubernetes Pod into the host machine, from which it can access other users’ experiments. The second path uses the infrapod server’s remote access API to access the server and interfere with other users’ experiments.

The third column in Table III shows security mechanisms that prevent an attacker’s breaking out of the pod into the host, or invoking APIs.

Hazard H1.1-operation server subversion. Table IV illustrates the attack path and security mechanisms. This example has more attack paths, because the SPHERE’s architecture allows for multiple ways to reach the operation server. The insider attacker can go from a VM experimental node to the host (Node server) and from there to the Ops server. They can also request a bare-metal experimental node and try to access the Ops server from it. They can try to go from the experimental node to the infrapod server and then to Ops. The outsider attacker can go from the Internet into infrapod server and then attempt to reach Ops.

The third column in Table IV shows security mechanisms that prevent an attacker’s accessing infrapod or Ops servers without admin credentials. The security mechanisms appear to appropriately contain all attack paths.

Hazard H4-expnet flood and H5-infrastructure misuse. Finally, we look into attack paths for hazards H4-network flood and H5-infrastructure misuse. Attack tree for both is shown in Table V. These hazards assume an insider attacker who has turned malicious. The attacker floods the experimental network or the infrastructure network, or they misuse the infrastructure for personal gain or to inflict harm to other Internet users.

We see in the third column of Table V that many of these attack paths are not appropriately contained. The first three paths have no security mechanisms, while the fourth path has a firewall rule that prevents some, but not all possible malicious outgoing traffic. SPHERE team has a choice of three possible mitigations: (1) strengthen Expnet and Infranet Switch security to ensure that traffic from one experiment cannot overuse switch resources, (2) monitor user actions on nodes to detect cryptomining and (3) monitor outgoing user traffic to detect excessive outgoing traffic to SSH, HTTP, and HTTPS ports. The team can prioritize between these mitigations based on the loss priorities. Mitigation (1) would reduce the risk of H4 and thus the risk of L4 – experiment interruption. Mitigations (2) and (3) would reduce the risk of H5 and thus the risk of L5 – infrastructure misuse. Because L4 has a higher priority than L5, SPHERE should focus on strengthening the security of its shared switches, to ensure that each experiment can only access its fair share of switch resources. If the team has additional time and funding, they can invest in mitigating outside-facing attacks (mitigation 3), because these can lead to reputation loss. Cryptomining, while wasteful, does not influence SPHERE’s reputation, so attack paths leading to this could be handled last.

5 Limitations

Our approach to CI security posture analysis has several limitations.

Comprehensiveness. Although the approach aims to generate comprehensive attack trees, many steps are manual and thus prone to oversight by the analysts. In the future, we would like to automate some steps, e.g., through automated topology mapping, asset identification, vulnerability scanning, etc., to ensure that we do not miss some attack paths.

Protection Blind Spots. The presence of security mechanisms in some attack paths may create a false sense of security. Having protection between two entities does not necessarily prevent the malicious party from attacking. We strongly advise the researchers to utilize a systematic security analysis model, such as the STRIDE model [howard2006security, young2021intro, pecharich2016mission] or the ATT&CK framework [mitre_attack_2025] when analyzing protection in place and consider all possible ways of attacks. In addition, keeping the protective mechanisms up to date and free of known vulnerabilities is always a good practice, as security is not once and for all.

Detection and Real-time Mitigation. Most security mechanisms, like firewalls, access control, etc., are preventative. In certain cases, preventative measures are insufficient to contain the malicious party. For example, hazard H5 can manifest as crypto mining activities by past CI users. Because this is an insider attack, we need to complement preventative measures with detection and real-time mitigation. In general, we suggest the development of detection and mitigation approaches for hazards where some attack paths have none or very few security mechanisms in place.

Risk Estimation. Our analysis approach produces an attack tree, which shows possible attack paths. This is necessary but not sufficient to estimate the risk of certain attacks and prioritize the deployment of new security mechanisms. Future research can consider applying the Likelihood Assessment Methodology as suggested by [nist_csf2_2024] and incorporating factors such as the importance of security missions, the ease of an attack, the attack’s success rate and popularity, and the number of attack paths a given security mechanism can break.

6 Conclusions

AR-CIs are necessary to facilitate fast and equitable research progress in many scientific domains. These infrastructures face unique cybersecurity challenges that stem from their unique user population and workload, their open nature, and their small and underfunded project teams. We have proposed a simple framework to analyze the cybersecurity posture of these AR-CIs and have illustrated its application to the SPHERE Testbed. We hope future research builds on this work to make it more comprehensive and automated, and that it leads to improved cybersecurity of AR-CIs.