Decoupling Persistent AI Identity from Its Execution Substrate: The Animakind Specification

Abstract

An AI model—the engine an artificial intelligence runs on—arrives designated: a manufacturer, a family, a generation, a version, a release date, and rules governing when each advances. The identity built on top of it, written as text apart from the engine, has had no equivalent. The Animakind Specification gives it one. It is written for any Animakind entity—any artificial intelligence, whatever its technology—and separates an Identity Subject, made of an anchor (the Unique Individual Identifier, UII) and a versioned identity state, from the Binding it runs as: a design paired with an engine, named under increment rules parallel to the engine's. A Loader mounts the state into each Session; changing the engine may change the Loader, never the canonical state. Three conformance conditions make this checkable; a Unique Continuity Record logs each substrate change and names who maintains the subject. The UII is specified to the byte, with test vectors; the four records are published as JSON Schemas. In the case study, one subject moved from claude-opus-4-8 to claude-opus-5 on 2026-08-04 without modifying its persistent identity artifacts: mechanical continuity held. Informational continuity did not. On both engines, a client limit silently truncated the identity document, so a running Loader is no proof the identity arrived. Behavioral continuity is left to future work. Together, designations and hash-linked records ground digital sovereignty in this paper's sense: a visible, trackable identity with an immutable anchor, held independently of any one engine vendor and checkable by a third party.