When AI Builds Itself, Who Remembers Why? Temperature Zero, Recursive Self-Improvement, and the Continuity Layer AI Safety Is Missing

This essay argues that the public debate around recursive self-improvement and AI safety is missing a continuity layer. Temperature zero can reduce token-level variation, retrieval can surface more information, and a pause can slow development, but none of these mechanisms automatically preserves the reasons, assumptions, evidence, decisions, and justified changes behind consequential AI-assisted work. The Mayorga Mnemosyne AI Continuity Framework™ proposes continuity as a neglected governance layer above inference, retrieval, logs, and emergency coordination. If AI systems begin helping build future AI systems, society will need a way to remember why important changes were accepted.