Assessing AI readiness in space systems

Space systems sit at the heart of modern critical infrastructure, with communications, navigation, weather services, logistics, and military operations all largely depending on them.

Assessing AI readiness in space systems

The threat landscape 

Space systems sit at the heart of modern critical infrastructure, with communications, navigation, weather services, logistics, and military operations all largely depending on them. Yet, their security has not kept pace with their criticality, and their exposure to cyber risk is growing just as the sector becomes more commercial and interconnected. The threat landscape spans all three architectural layers: the ground segment (ground stations and control centres), the link segment (the radio frequency channel binding them), and the space segment (satellites) – each presenting distinct and compounding vulnerabilities. 

The ground segment 

Ground infrastructure – TT&C (Telemetry, Tracking and Command) stations, control centres, and specifically network operation centres – runs on standard IT and operational technology, exposing it to the full conventional threat spectrum (e.g., remote access services, misconfigurations, unpatched software, phishing, credential theft, third-party compromise, etc.). ENISA notes that internet-connected ground assets are among the lowest-sophistication entry points into a space system for attackers. 

The 2022 Viasat KA-SAT attack remains the defining case study: Russian-attributed actors exploited a misconfigured VPN appliance in the ground management network and deployed a wiper malware, rendering tens of thousands of satellite modems across Europe inoperable, without ever targeting a single satellite. The cascade effects extended to Ukrainian military communications and approximately 5,800 Enercon wind turbines in Germany. 

Recent reporting suggests that the broader trend is worsening. The Space ISAC Q3 2025 Public Threat Report documents a rise in geopolitically motivated cyber-attacks in 2025, with named threat actors including China-nexus Salt Typhoon (confirmed to have targeted a US satellite communications provider), Iranian APT group UNC159/Subtle Snail (deploying tailored malware against European aerospace and defence organisations), and the newly identified RedNovember cluster, extending operations specifically to space organisations. The report also points to supply chain vulnerabilities in widely used enterprise products such as Citrix, SharePoint, SAP NetWeaver, and SonicWall as recurring entry points. 

The link segment 

The traditional radio frequency channel is a broadcast medium whose inherent characteristics increase its threat exposure as well as defensive complexity compared to terrestrial networks. 

The Space ISAC’s Q3 2025 notes that publicly available GNSS reports of jamming and spoofing exceeded 2000 per month, with multiple commercial aviation flights routinely encountering disruptions severe enough to affect landing procedures, and maritime operators in the Red Sea and Strait of Hormuz regularly reporting GNSS impacts to navigation and speed data. 

The space segment 

The fundamental paradox of satellites is that they cannot be serviced once in orbit, making design-stage weaknesses especially important. ENISA notes a wide range of relevant threats at this layer, including unauthorised access, malicious command injection, malware, software vulnerabilities, and supply-chain compromise. 

In 2023, researchers from Ruhr University Bochum showed how concrete some of these risks are. In a study of three satellites, including ESA’s OPS-SAT and Estonia’s first satellite ESTCube-1, they found that weaknesses such as unencrypted command channels and standard software vulnerabilities could be used to gain control of core onboard systems and run malicious code. The point was less the sophistication of the attack than the ordinariness of the flaws: in some cases, space systems were exposed to the same kinds of IT vulnerabilities seen on Earth. 

Supply chain risk is also a recurring concern. Satellite programmes depend on long chains of contractors and on commercial off-the-shelf components, which can introduce pre-launch vulnerabilities, such as malicious or vulnerable code that remains dormant through testing and activate only once in orbit. And since satellites within a constellation often share software architectures, a single compromised component can affect an entire fleet simultaneously. 

These three layers are operationally tightly connected. A compromise in the ground segment can disrupt services across an entire satellite network. Interference in the link segment can corrupt the data on which both operators and users rely. Weaknesses in onboard software become more serious when the asset is physically unreachable. ENISA’s comparative analysis concludes that space systems carry higher threat exposure than comparable terrestrial networks due to their reliance on wireless links, and the difficulty of response and recovery given by their location. 

How AI is being used for space cybersecurity 

AI is beginning to play a real role in space cybersecurity, but its use is uneven across the architecture. In practice, most current applications are focused on specific operational challenges such as anomaly detection and threat intelligence analysis. 

The defence of space systems is gradually shifting from reactive patching to predictive monitoring, driven by machine learning. However, applying AI to space cybersecurity is not as simple as porting over enterprise IT software, as its deployment is dictated by the physical and computational realities of each segment. The strongest activity is on the ground, where compute, storage, and software update cycles are less constrained, while onboard use in satellites remains far more limited. 

The ground segment and computing 

Because ground stations have virtually unlimited power and processing capacity, this is where most defensive AI currently lives. Ground systems generate large volumes of telemetry, network logs, command histories, and system health data, which makes them suitable for machine-learning models trained to identify deviations from normal behaviour. This means that the AI can flag unusual command sequences, changes in power or thermal profiles, access patterns, or combinations of events that would be difficult for a human analyst to correlate in real time. 

A major hurdle for these systems, however, is a lack of training data. Real cyber incidents in space systems are relatively rare in public reporting and often classified. This is one reason the sector is turning to generative AI and digital twins: by simulating satellite networks, researchers can generate high-fidelity synthetic data representing various attack scenarios, used to produce training data for defensive models without exposing real operational systems. 

The link segment and the processing of RF signal 

To protect the communication links, AI is increasingly applied to radio frequency (RF) signal processing. Machine learning excels at pattern recognition, making it highly effective at monitoring the RF spectrum for anomalies. AI models are being trained to detect the subtle, early signatures of GNSS spoofing or jamming – distinguishing between legitimate satellite broadcasts and malicious signals attempting to hijack a receiver – before human operators even notice the interference. 

The space segment and the limits of edge AI 

The ultimate goal is to put defensive AI directly onboard the satellite – often called Edge AI for spacecraft. If a spacecraft loses contact with the ground due to jamming or a ground station breach, on-board AI could theoretically detect the intrusion and sever compromised uplinks autonomously. 

In practice, this is severely constrained by SWaP (Size, Weight, and Power). Because space-rated hardware must survive heavy radiation and temperature extremes, many operational satellites run on processors roughly equivalent to 1990s-era desktop computers. Consequently, current on-board AI is limited to task-specific, lightweight anomaly detection models that monitor basic system health and command execution, rather than performing full autonomous cyber defence. 

The gaps 

While AI may improve space cybersecurity, current implementations face severe operational and physical bottlenecks. 

  1. Data scarcity: as mentioned above, machine learning models need large volumes of training data. However, as real-world cyber incidents in space are rarely shared due to classification barriers, researchers are trying to generate synthetic datasets by using digital twins. Nevertheless, as adversarial tactics evolve, simulated environments can drift away from operational reality.
  2. Space hardware: ideally, the most resilient defensive posture would involve placing AI directly on the satellite, allowing it to detect threats even if communications with the ground are disrupted, but onboard processors are limited by SWaP and hardening requirements. Therefore, most of the computational work for AI defence remains in the ground segment, which means AI does not yet remove the system’s dependence on its most exposed layer.
  3. AI attack surface: deploying machine learning models in space architectures makes the AI itself a target of adversarial ML tactics, such as training data poisoning and model evasion, where an attacker manipulates either the data used to train a model or the inputs it sees in operation.

The space-AI maturity matrix 

The current state of AI in space cybersecurity is defined by a structural paradox: defensive AI capability is inversely proportional to system isolation. 

The architectural layer that most requires autonomous defence – the space segment, operating beyond physical reach and vulnerable to communication denial – is the one least capable of supporting it computationally. Conversely, the layer with the highest computational power – the ground segment – is already the system’s most accessible vulnerability. Using advanced AI on the ground certainly improves the network’s analytics, but it also concentrates reliance on its most exposed targets. 

This dynamic yields a clear maturity matrix for assessing AI readiness stages across the architecture: 

  • Ground segment: operational. Supported by terrestrial computing, defensive AI applications are actively deployed today.
  • Link segment: emerging. Applications driven by machine learning for signal processing are rapidly transitioning from experimental to practical.
  • Space segment: constrained. Autonomous onboard cyber defence remains blocked by the physical limitations (SWaP) of space-rated hardware.

In short, AI is currently a powerful diagnostic tool, but until space-rated processors become significantly more powerful, a fully self-defending satellite will remain a hardware problem, not a software one.


Francesco Macci holds an Erasmus Mundus International Master’s Degree in Security, Intelligence, and Strategic Studies (IMSISS), a master’s degree in Leadership for International Relations and Made in Italy, and has completed an advanced training course in Space Institutions and Policies with a scholarship from the Italian Space Agency (ASI). For the Moroccan Institute for Policy Analysis (MIPA), he published “The Growth of the Moroccan Military Air Power” (2023). He coordinated the Security & Defense Working Group of the European Student Think Tank (EST); he has worked as an intelligence analyst in the private sector, and currently serves as a Country Security Manager.


Sources
ENISA, LEO Satcom Cybersecurity Assessment, 2024
J. Willbold et al., “Space Odyssey: An Experimental Software Security Analysis of Satellites,” in 2023 IEEE Symposium on Security and Privacy (SP), San Francisco, CA, USA, 2023, pp. 1-19, doi: 10.1109/SP46215.2023.10351029
Space ISAC AI/ML COI, Generative AI for Data-Driven Space Operations, 2025
Space ISAC and OASIS, SATIS Announcement, 2024
Space ISAC, Developing AI for Space Cybersecurity, 2025
Space ISAC, Q3 2025 Public Report, 2025