Large Language Models (LLMs) offer a compelling promise for satellite communications (SATCOM): supporting operators in complex, time-critical environments by turning vast amounts of data into actionable insight.
In satellite operations, operators are often confronted with large volumes of documentation, procedures, logs, and telemetry data. Generative AI can help extract exactly the information that is relevant for a specific situation or an ongoing incident — and do so within seconds instead of hours.
Beyond that, LLMs can summarize logs into a clear, up-to-date operational picture, assist in root-cause analysis, and even help draft scripts or command sequences to support recovery actions or routine tasks. Used correctly, such tools could reduce workload, speed up response times, and improve overall situational awareness.
However, this potential comes with serious risks.
LLMs are not inherently trustworthy systems. They can be attacked, manipulated, or misused. They may produce confident but incorrect outputs, make up information, or fail to follow instructions consistently, especially in complex or adversarial settings. In safety-critical domains like SATCOM, these weaknesses are not just theoretical concerns; they can translate directly into operational and security risks.
This is where our project comes in:
- Demonstration: We are building an AI Helper for Satellite Operations to showcase what LLMs can realistically contribute in an operational context.
- Risk Analysis: We deliberately attack this AI Helper to expose concrete vulnerabilities, attack vectors, and critical failure modes and to understand which factors make them more likely.
- Mitigation: We explore and evaluate measures to harden LLM-based applications against attacks and critical errors, with the goal of making such systems safer for real-world use.
Generative AI in SATCOM is neither pure hype nor a solved problem. It is a powerful tool that must be approached with caution, technical rigor, and a strong security mindset.
Stay tuned: first results will be coming soon. 🚀.