Adversarial Machine Learning Threats to Spacecraft - Paper

Aug 2, 2026

Uploaded by Asher

Spacecraft operate as early autonomous systems whose ability to function without human-in-the-loop has enabled major missions. As autonomy expands, space vehicles face growing exposure to attacks that target and disrupt probabilistic autonomous processes, especially through adversarial machine learning (AML). This work presents an AML threat taxonomy for spacecraft, then demonstrates AML attack execution via experimental simulations using NASA’s Core Flight System (cFS) and OnAIR platform. Results emphasize integrating AML-focused security measures for autonomous spacecraft.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages11

Reading Time28 min

AccessFree to download

Introduction

Space as a contested security domain

AI integration and emerging AML vulnerabilities

Motivation and gap in risk assessment

Background and Prior Art

Adversarial machine learning overview

Adversarial Machine Learning Threats to Spacecraft 1st Rajiv Thummala Sibley School of Mech. and Aerospace Engineering Cornell University Ithaca, NY USA[rkt34@cornell.edu](rkt34@cornell.edu) 2nd Shristi Sharma Department of Computer Science The University of North Carolina at Chapel Hill Chapel Hill, NC [ssharma@unc.edu](ssharma@unc.edu) arXiv :2405 .08834v1 [ cs .LG] 14 May 2024 3rd Matteo Calabrese Department of Systems Engineering Cornell University Ithaca, NY USA [mc2884@cornell.edu](mc2884@cornell.edu) 4th Gregory Falco Sibley School of Mech. and Aerospace Engineering Cornell University Ithaca, NY USA [gfalco@cornell.edu](gfalco@cornell.edu) Abstract—Spacecraft are among the earliest autonomous systems. Their ability to function without a human in the loop have afforded some of humanity’s grandest achievements. As reliance on autonomy grows, space vehicles will become increasingly vulnerable to attacks designed to disrupt autonomous processesespecially probabilistic ones based on machine learning. This paper aims to elucidate and demonstrate the threats that adversarial machine learning (AML) capabilities pose to spacecraft. First, an AML threat taxonomy for spacecraft is introduced. Next, we demonstrate the execution of AML attacks against spacecraft through experimental simulations using NASA’s Core Flight System (cFS) and NASA’s On-board Artificial Intelligence Research (OnAIR) Platform. Our findings highlight the imperative for incorporating AML-focused security measures in spacecraft that engage autonomy. I. INTRODUCTION The space domain has undergone a significant paradigm shift, evolving from a predominantly exploratory and scientific domain into a strategic and contested sphere, integral to national security and defense postures globally. This transformation is underscored by the recognition of space as a critical theater for military operations, where the assets deployed—ranging from communication and navigation satellites to surveillance and reconnaissance platforms—constitute pivotal elements in the infrastructure of modern warfare. As nations increasingly depend on these assets for both strategic advantage and operational capabilities, the imperative to safeguard them from a spectrum of threats has never been more paramount. Concurrently, the integration of artificial intelligence (AI) within space systems has been accelerating, driven by the need for enhanced data processing, autonomous decision-making capabilities, and improved operational efficiency. This reliance on AI systems marks a significant evolution in spacecraft design and functionality, enabling more sophisticated missions and responsive space operations. However, the incorporation of AI also introduces new vulnerabilities, particularly in the context of adversarial machine learning (AML) . AML techniques, which are designed to manipulate or deceive AI models through carefully crafted inputs, represent a burgeoning threat vector that spacecraft systems are ill-prepared to counteract. The hardening of spacecraft against cyber threats, while nascent, has been an area of growing focus within the aerospace defense sector. Yet, the threats posed by AML specific to spacecraft have not been adequately explored or addressed, leaving a critical gap in risk, threat, and vulnerability assessments for spacecraft engineers. The sophistication and adaptability of AML attacks exacerbate this oversight, as traditional cybersecurity measures prove insufficient against threats that are designed to exploit the inherent vulnerabilities of AI systems. This discrepancy underscores the urgent need for a concerted effort to understand the unique implications of AML in the space domain, to inform spacecraft engineers and designers in effectively diagnosing and mitigating these emerging threats. Accordingly, this paper details a taxonomy of AML threats to spacecraft, with emphasis on the dynamics between mission-specific objectives, the operational

FAQs

Why are spacecraft increasingly vulnerable to adversarial machine learning attacks?

Spacecraft autonomy and probabilistic decision processes rely heavily on ML components, which can be manipulated by carefully crafted inputs that deceive or corrupt model behavior.

What contributions does the paper make to address AML threats to spacecraft?

It introduces an AML threat taxonomy tailored to spacecraft and provides experimental demonstrations of AML attacks through spacecraft simulations.

How are AML attacks demonstrated in the paper?

The authors run experimental simulations using NASA’s Core Flight System (cFS) and the NASA On-board Artificial Intelligence Research (OnAIR) platform to execute adversarial attacks against spacecraft models.

Aug 2, 2026

Uploaded by Asher