The Kretauer–Kelly Bridge Model: Human-in-the-Loop Governance for Limiting Recursive Progression under Uncertainty

English Description The Kretauer-Kelly-Bridge Model (KKB) is proposed as a theoretical governance framework for human-accountable decisions about expanding the operational scope of autonomous AI systems under uncertainty. Its focus is not on calculating optimal system performance, but on reasoned human authorization of a specifically described progression step. The model addresses whether, and under which conditions, a system should receive additional tasks, permissions, or opportunities for deployment. The decision architecture connects three core dimensions: system state, progression impetus, and an accountable human decision authority. Decision-relevant uncertainties, available evidence, possible alternatives, and the consequences of non-progression are explicitly considered. The accompanying decision protocol distinguishes authorization, refusal, and deferral. Authorization applies only to the specified scope and validity period; deferral grants neither provisional permission nor partial authorization. Working version 1.1 distinguishes the general model, its context-specific organizational configuration as a KKB instance, and actual implementation. The human decision point precedes implementation of a covered expansion. Existing authorization procedures, risk assessments, and safety assessments are incorporated rather than replaced. The approach is positioned as a possible process component within broader AI governance; it does not claim that comparable decision procedures are generally absent. The additional rule proposals R1–R4 specify how to address missing decision authority and conflicts of interest, cumulative expansions, the limits of conditional authorization, and the distinction between renewed review and authorization expiry. These additions are explicitly marked in this working version as proposals pending substantive author approval. Five empirically testable hypotheses concern traceability of authorization, treatment of decision-relevant uncertainty, actual compliance with authorized scope, effective human decision authority, and additional value at acceptable cost. The proposed research programme includes comprehensibility and consistency assessment, controlled vignette comparisons, and subsequent context-specific field evaluation. It distinguishes possible effects of the protocol from the effects of additional resources, mandates, or technical safeguards. The publication is a theoretical model proposal and working paper, not an empirical validation study. It reports neither a safety demonstration nor a demonstrated advantage over existing procedures. Human authorization and complete documentation do not guarantee sound decisions or effective limitation. Whether KKB provides additional value, has no effect, or creates disadvantages remains a question for the proposed evaluation.