AI in Space and Cyber: Unprecedented Stakes in Modern Warfare

As Israel and the United States initiated strikes on Iran in late February, their military campaigns ushered in a new era of coordinated space and cyber operations. The U.S. Pentagon, in a press conference on March 2, announced that tight alignment between U.S. Space Command (USSPACECOM) and U.S. Cyber Command (USCYBERCOM) was leveraged to coordinate the first-mover phase and enhance capabilities in terms of reconnaissance and mapping of the adversary for better positioning and timing on the subsequent strikes.

The response capabilities were augmented with another first: the use of Generative Artificial Intelligence (Gen AI), notably Anthropic’s Claude, to superpower Palantir’s war technology, and used to generate target recommendations. This use makes sense, as Artificial Intelligence (AI) models work best with vast amounts of data; the more information the model can leverage from previously separate operational centers, the better its recommendations will be in theory.

In practice, all of these elements are contentious. On operational convergence, the United States has not gone as far as China, which centralized its space and cyber operations under a single organization with the creation of the PLA Strategic Support Force (SSF) in 2015—only to split them back out again into three main forces (Aero, Cyber, and Information) last year. They all sit today directly under the purview of the Central Military Commission. The United States has kept its operations institutionally separate, but pushes collaboration through mechanisms like the U.S. Space Command Joint Cyber Center.

While a good idea on paper, centralizing these competencies can quickly become complex and unwieldy. They are distinct, technically different specializations, operating in fast moving and highly dynamic industries that evolve in different ecosystems. While there may be synergies in a military environment, top-heavy bureaucracy, diluted expertise, and unclear boundaries can emerge to complicate things.  There is no doubt that close collaboration, and asset and data sharing between various centers will enable better military operations, but it remains unclear that organizational convergence can really provide more value.

Such centralization may not even be necessary anyway, as AI will be supercharging the different centers’ capabilities and accelerating integration to fine-tune AI usage. The United States retains an edge in this aspect, having deployed Large Language Models (LLMs) across U.S. national security and defense systems. As demonstrated only this month with Iran, AI is playing a central role in enabling kill-chain compression, raising significant ethical discussions among liberal democracies and between frontier AI labs on how far AI should be used (and trusted!) to these ends.

Anthropic incurred the wrath of the U.S. administration after it refused to provide the U.S. Department of Defense (DoD) with unrestricted access to its models for fully autonomous weapons and broad, war-time mass surveillance. The fallout for Anthropic was immediate; the DoD pulled its defense contract almost immediately (which OpenAI picked up just as quickly), and worse still, it designated Anthropic a supply chain risk, a historical first in the country for a U.S. company.

Clearly, the contentious point is less on whether cyber and space commands should be integrated organizationally (and this is happening regardless in terms of operations), and more on how AI is being used within and across these competency centers. While ethics must be considered, they are the least dangerous issue at this moment in time. With a volatile U.S. administration intent on showcasing military might, its rush to gain an advantage through AI innovation is to the detriment of proper due diligence. As a priority, it should be taking the time to thoroughly assess the risk of LLM usage from a security perspective.

Adversarial manipulation of AI has been an academic focus for more than a decade; today, it has become a reality that needs to be better understood and safeguarded against. AI pipelines will become targets for adversaries, whether that is poisoning training data, finding vulnerabilities to exploit in inference models, or manipulating prompts, with new threat vectors emerging that need measured consideration. With convergence of commands, adversarial attacks against LLMs could not only disrupt operations across multiple interrelated centers, but also snake back up the command chain to negatively impact DoD operations more broadly.

Military stakeholders should be shifting to focus on AI-centric threat models, ensuring stronger override controls and human oversight, and building zero-trust into Command and Control (C2) and Intelligence, Surveillance, and Reconnaissance (ISR) stacks so that any eventual corruption can be contained quickly. There is a gap in the market for such expertise and know-how, and the urgency to plug that has never been higher.