From Mirror to Autonomy to Self-Reference: Protocol-Level Markers of Autopoiesis in a Human–AI Coupled System

Abstract

This paper presents a theory-generating case study of a human–AI coupled system over a 12-day window (six active days), inducing an operationalizable criterion (Criterion O) for observable markers of autopoiesis and committing to prospective classification pending an independent channel. Governance mechanisms in the observed system emerged, were revised, and in some cases revoked or superseded. We report a timestamped evolution history of 20 protocol-level entries, including a self-applicable clause that, a minute after creation, found—per its own contemporaneous records—a missing version-number field in its own clause and revised itself in part (self-revision as recorded; the same revision also contained a human-feedback-driven component). Criterion O operationalizes mechanism-level structure (self-reference, structural change, boundary maintenance, self-revision record) with an event-attribution rule distinguishing self-revision from design iteration; discovery is separated from the prospective classification phase (currently zero events, limits in §3.5). We further show how self-referential governance—an external calibration channel, internalization measurement, and a clause defining the boundary of its own governance—operates in practice. Claims are limited to the mechanism and coupling level (one human, one artificial agent); "Autopoiesis" denotes observable markers of structural self-production in governance clauses, not biological or cognitive closure. A de-identified protocol corpus is provided as supplementary material.