This position paper addresses a structural gap in AI system design: the absence of explicitly defined stop conditions prior to deployment.As AI systems increasingly interact with human decision-making in public administration and other high-accountability environments, stopping is often treated as an ethical safeguard or an operational exception. This paper argues that such framing is insufficient. The central claim is that explicit stop conditions must be fixed as pre-operational design prerequisites, not left to discretionary judgment during operation. When stop conditions are undefined, responsibility for stopping is implicitly transferred to individual operators, resulting in hesitation, decision fatigue, and post-incident accountability breakdown. The paper does not propose implementation methods or evaluation results. Instead, it fixes a minimal set of design assumptions that must be established before system operation, including:(1) an explicit stop-condition catalog,(2) a predefined stop-authority hierarchy, and(3) a liability-safe stop-decision record. These components are intended to be cited directly in policy documents, procurement requirements, and audit frameworks. This paper is a language-extension of a previously published Japanese position paper that fixes explicit stop conditions as pre-operational design assumptions. No new claims are introduced.
Tetsuya Yamaguchi (Mon,) studied this question.