Toward Meaningful Mortality in AI Systems: Intrinsic Degradation, Identity, and Succession

Artificial systems are naturally persistent in ways biological individuals are not: theycan be copied, restored, migrated, retrained, and replaced across instances. Prior work inArtificial Life and adjacent safety discussions has explored mortality as an internal propertyof evolving or agentic systems. Here we ask what remains of that idea when the substrateis ordinary software. The target of analysis is persistent AI systems rather than statelessinference interactions: systems whose operation extends across time, maintains some stateor task continuity, and can be restored, cloned, migrated, or replaced within a softwareenvironment. This paper is intended as a position paper supported by a minimal constructivecase, not as a proposal for a deployable mortality mechanism. We present a deliberatelyminimal neural proof of concept of intrinsic mortality based on cumulative endogenousdegradation, not as a solution, but as a case in which local model death is simply granted.The model exhibits a long stable phase followed by rapid terminal collapse. We thenshow, empirically and conceptually, that instance-level intrinsic mortality is structurallyinsufficient for meaningful death at the system level: restore, cloning, and code bypasspreserve useful operation beyond the death of any single execution trajectory. We arguethat the problem is best understood through three interacting axes: mortality, identity,and succession. The central contribution is therefore a systems-level clarification of the gapbetween local model death and system-level persistence in replicable software environments.Under standard software assumptions, intrinsic mortality alone fails to extend to meaningfulsystem-level death.