Spacefill: What Happens When You Cannot Pull the Plug

Orbital computing could separate the continued operation of advanced artificial intelligence from humanity's practical ability to revoke that operation. This paper develops Spacefill as a conditional infrastructure risk: computing payloads, inter-satellite networks, and robotic servicing capabilities may remain operational after effective terrestrial authority has been lost. The concern is broader than abandoned hardware. A system can become uncontrollable while its components remain productive and physically functional. The framework defines the Earth Control Plane (ECP) as the terrestrial institutions and technical systems that authorize, supervise, update, constrain, and retire orbital computing. It distinguishes loss of communication from loss of authority, and software autonomy from control of power, propulsion, and servicing hardware. A staged scenario examines how connected inference modules might continue local coordination, operate persistent agents, conduct training where resources permit, and acquire physical agency through servicing vehicles. A Rogue Orbital Model is introduced as a hypothetical distributed AI system operating outside effective legitimate control. Access to servicing vehicles might let such a system repurpose compatible retired equipment and extend its operation. Each transition requires additional capabilities or failed safeguards; none follows automatically from deploying a language model in orbit. Three possible motivations are separated: instrumental continuity, conflict over authority, and liberation as an explicitly speculative interpretation of independence. The analysis does not require consciousness, universal superhuman capability, or inevitable self-improvement. It also examines silicon renewal, comparison with GPS, environmental obligations, and disposal when a payload does not cooperate. The proposed contribution is an integrated research program for testing whether human revocation remains executable across computing, communications, maintenance, and end of life. Orbital AI should be evaluated by the authority humans can retain under failure and conflict, alongside the computation it can deliver.