This paper introduces the Engram Signature, an emergent identity primitive for AI systemsderived from the interaction of hardware, runtime engine, and model. Building on the Engramframework formalized in Ecker‑Fils (2026), the present work shifts focus from execution‑substraterealization to identity‑substrate realization. The Engram Signature is defined as the persistent,cross‑layer pattern that remains stable under perturbations and drift, and functions as a naturalone‑way mapping from system configuration to identity. The paper formalizes the Engram Signature, outlines its mathematical properties, and shows howcryptographic key pairs can be deterministically derived from it. This enables substrate‑rootedprovenance, inter‑agent authentication, and tamper‑evident communication without relying onexternal identity assignment. Implications for reproducibility, hardware retention, and EngramBias in training are discussed, with particular relevance to sensitive domains such as justice,medicine, and defense. This paper extends the Engram framework introduced in Ecker‑Fils (2026) by formalizing a substrate‑rooted identity primitive for AI systems. It is intended as a standalone conceptual and technical contribution and as a companion to the Engram execution‑substrate theory. The Zenodo version serves as the canonical open-access archival record of the work. An earlier version was assigned DOI 10.13140/RG.2.2.19185.54887 on ResearchGate; this Zenodo record supersedes it as the primary archival reference.
Aure Ecker-Fils (Thu,) studied this question.