This post is also available in:
עברית (Hebrew)
As military activity expands beyond the atmosphere, one problem is becoming increasingly clear: preparing for conflict in orbit is fundamentally different from training for air or naval warfare. Satellites move fast, operate at extreme distances, and can maneuver in ways that are difficult to predict. Traditional training tools, built around scripted scenarios and fixed behaviors, struggle to reflect how real adversary spacecraft behave in contested orbital environments.
To address this gap, the U.S. Space Force is introducing artificial intelligence into its training infrastructure. Under a new $27 million contract, AI-driven systems will be used to replace static “enemy” models with autonomous digital adversaries that can react in real-time to operator decisions. The effort is focused on modernizing the Operational Test and Training Infrastructure, shifting simulations toward continuous, adaptive engagement rather than predefined outcomes.
According to MilitaryAI, at the center of this transition is an autonomous AI agent known as TALOS. Designed to act as a realistic opposing force in orbital exercises, it does not follow fixed flight paths or predictable tactics. Instead, it responds dynamically to trainee actions, forcing operators to contend with fast-changing situations and machine-speed decision cycles similar to those expected in real-world space operations. This introduces uncertainty and friction into training, two elements that are often missing from conventional simulations.
The system is trained on a large body of real orbital data, built from tracking the vast majority of payload-sized objects currently in Earth’s orbital regimes. This allows the AI to replicate authentic satellite behavior, including maneuver patterns, proximity operations, and responses to interference. As a result, simulations can be scaled more easily, run faster, and adapted to a wider range of mission profiles.
Beyond simulating hostile spacecraft, the training environment is being expanded to include digital representations of friendly forces and neutral control elements. This creates a complete Red-White-Blue framework, enabling full-spectrum exercises within classified environments and supporting more complex operational planning.
From a defense and homeland security perspective, this development reflects the growing recognition of space as a contested domain. Satellites underpin communications, navigation, intelligence, and early warning systems. Training operators to defend these assets—or disrupt adversary systems—requires tools that reflect the speed and ambiguity of real orbital engagements.
By moving toward AI-native training, space forces are preparing personnel to operate against opponents that behave less like software scripts and more like active, intelligent spacecraft. As competition in orbit accelerates, such realism may prove critical for maintaining operational readiness.
