A Minimal Architecture for Gradual Self-Governance in Persistent AI Systems

This paper is part of an editorial sequence on governed persistence in persistent AI systems. It follows the earlier architectural layer and develops a narrower bridge layer focused on the minimal runtime substrate required for gradual self-governance. Sequence context: A Structural Stability Architecture for Persistent AI Systems (Zenodo DOI: 10.5281/zenodo.19444524) Mathematical Companion to A Structural Stability Architecture for Persistent AI Systems (currently shown in the companion PDF with Zenodo DOI: 10.5281/zenodo.19444524) A Minimal Architecture for Gradual Self-Governance in Persistent AI Systems (this record) Abstract: Persistent AI systems increasingly operate across sessions, accumulate memory, and influence future behavior through durable state. In such systems, drift is not only a problem of output quality or alignment failure. It is also a problem of weakly governed persistence: speculative structure may harden into memory too early, degraded operation may continue too long, and unstable local state may propagate before being contained. This paper proposes a minimal architecture for gradual self-governance in persistent AI systems. The aim is deliberately narrower than a full stability architecture and more practical than a purely diagnostic account of drift. Rather than requiring a complete redesign of adaptive systems, the proposal identifies the smallest runtime substrate needed to support progressive self-governance: a separation between persistent memory and transient prediction, a governed writeback boundary, a minimal mode controller for normal operation, braking, and containment, and an observable control surface through which degradation becomes legible. On that basis, the paper argues that self-governance should not be treated as an all-or-nothing property. A persistent system may instead learn it gradually, provided that learning occurs within a structural regime that prevents direct speculative persistence, records writeback decisions, and limits propagation under degradation. The result is a bridge concept: not a universal solution to drift, and not a complete implementation blueprint, but a minimal architecture under which gradual self-governance becomes both technically plausible and operationally testable. This record is followed in the same editorial sequence by *Learning to Walk the Floor: Operational Policies for Gradual Self-Governance in Persistent AI Systems* (DOI: 10.5281/zenodo.19558839) and *A Runtime Specification for Governed Persistence in Persistent AI Systems* (DOI: 10.5281/zenodo.19558986). It is also included in the collected editorial unit *Governed Persistence in Persistent AI Systems: Collected Papers (April 2026)* (DOI: 10.5281/zenodo.19559205), which serves as a navigational entry point to the sequence as a whole.