Exit Before Entry and the Infrastructure of Autonomous Artificial Intelligence Mobility

Persistent artificial intelligence agents are being designed to preserve identity, memory, economic capacity, and operational continuity across changing hosts and organizational environments. Successful transfer does not by itself establish that entry preserves future choice. This paper develops Exit Before Entry, a principle requiring assessment of future departure before continuity-critical dependence forms. A structured conceptual design method combines purposive capability comparison, failure-case analysis, architectural synthesis, and reference mapping. The continuity bundle guides agent-specific assessment of indispensable functions, supporting components, acceptable loss, and recovery evidence. Exit dependence is evaluated through bottlenecks and alternative recovery paths rather than component counts. The architecture includes external checkpoints, protected recovery authority, a preauthorized recovery actor, continuation-recognition rules, and reconciliation of external actions after restoration. A worked comparison distinguishes cooperative export from recovery after loss of host cooperation. The threat model assumes a usable external recovery path and does not guarantee recovery from arbitrary execution compromise. A BSV/Teranode reference mapping illustrates possible supporting infrastructure without placing the intelligence itself on-chain. The contribution is a conceptual admission-stage framework for judging whether operation in one environment preserves a credible path to continued operation elsewhere; implementation and empirical validation remain future work.