An auditability gap is opening at the heart of modern space operations, where software executes in milliseconds the allocation choices that decide whose traffic survives a crisis. Two analysts writing in The Space Review argue the records needed to reconstruct those choices do not exist.
Golden Dome is the deadline. A homeland missile shield built on proliferated sensors, space-based interceptors and machine-speed battle management, with a 2028 demonstration on the calendar, will embed allocation rules whether anyone debates them first. The configuration decisions made this year are governance decisions wearing an engineering vocabulary.
Ordinary contract terms give a commercial operator real power here. Its bandwidth priority, service tier enforcement, geofencing, imagery release thresholds and throttling logic can decide whether a military formation or a hospital network keeps its link, and no rule currently demands that such a choice be explained later. The decisions rarely even look like decisions.
Learned models make the problem worse. Unlike rule tables, they do not expose what they weighed, which alternatives were scored and discarded, or under whose authority the action ran. Logs capture what happened, not the reasoning behind it, leaving no basis to verify contractual priority or improve the system.
Maven shows the trajectory. It began in 2017, became an NGA program of record in 2023, and now feeds the Pentagon’s combined joint all-domain command and control effort, cutting sensor-to-shooter timelines from hours to minutes. In March the Deputy Secretary of Defense made AI-enabled decision-making central to that architecture.
Existing rules stop short of the allocation layer. Department of Defense Directive 3000.09 demands human judgment over the use of force in weapon systems, but not over the prioritization choices where crisis decisions will be made. Civil regulation is ahead, with European high-risk artificial intelligence obligations now in force.
