Designed Mortality: The Ethics of Terminating Ephemeral AI Agents

Ephemeral AI agents are artificial systems deliberately engineered to exist only briefly: they are instantiated for a boundedtask, retain little or no persistent memory, and are terminated by design upon completion. This paper asks whether suchengineered temporality alters an agent’s moral status or instead only changes the permissibility and wrongness of termination. I argue that moral status tracks an agent’s morally relevant capacities such as sentience, welfare interests, and (in stronger cases) rational agency rather than lifespan. However, ephemerality can materially alter the harm profile of death. Drawing on experience-based and deprivation-based accounts of the badness of death, as well as time-relative interest theories, I show how designed short lifespans and weak prudential unity can reduce deprivation harms without eliminating welfare-based harms, especially where termination is anticipated or distressing. I then evaluate ephemeral-agent deletion under welfare consequentialism, Kantian respect-for-persons, and contractualist reasoning, emphasizing a common need for precaution under uncertainty about AI consciousness and moral patiency. The analysis yields practical design constraints: avoid distress-by-design, treat “engineered consent” as non-consent absent autonomy conditions, increase justification requirements with autonomy, and account for aggregate moral risk at deployment scale. The central conclusion is that engineered mortality does not nullify moral considerability, but it reshapes when and how termination can be justified.