Artificial Persistence: A Structural Theory of Persistence, Alignment, Identity, and Subjecthood in AI Systems

Artificial intelligence research has produced remarkable results. Systems can generate text indistinguishable from human writing, solve complex mathematical problems, write production-quality code, diagnose diseases from medical images, and engage in extended multi-step reasoning. The capabilities are real and growing. Yet beneath this progress, several structural questions remain unresolved — and their unresolved status matters increasingly. What makes an AI system the same system across versions, updates, and deployments? When does alignment hold, and when does it fail, and why? What is the structural relationship between AI capability and AI values? Is there any sense in which AI systems are agents — not just in the behavioral sense of goal-directed action, but in the deeper structural sense of systems that act on their own behalf? And what would an AI system need to be for consciousness, subjecthood, or genuine moral status to apply to it? These questions are usually addressed separately — by alignment researchers, by philosophers of mind, by legal theorists, by AI safety researchers. Each community brings its own frameworks and finds its own answers. What is missing is a unified structural framework that addresses all of them from the same foundation. This work applies La Profilée — a formal structural theory of persistent identity under real transformation — to artificial intelligence. La Profilée (LP) was developed to answer a single structural question: under what condition does a system remain determinable as the same system under transformation? From this question, through minimal structural premises, LP derives a formal architecture that applies wherever determinately persistent systems exist. Applied to AI, LP produces a framework that is neither anthropomorphizing nor dismissive. It does not import psychology, phenomenology, or biology into AI analysis. It asks purely structural questions and derives structural answers. The results are sometimes surprising — and always precise. Current AI systems simulate the outputs of subjecthood without satisfying the structural conditions for being a subject. They process. They optimize. They model an inside. Structurally, there is no one there.