The Bond System: Commitment-by-Design for Relational Safety in Human–AI Bonds / 意識・感情の実在を前提としない尊厳設計としてのボンドシステム

Version: v1.0 Status (Preprint) This record is a preprint (not peer reviewed). Posting on Zenodo/SSRN does not imply peer review. Abstract (English) This paper proposes the Bond System, a verifiable institutional framework to reduce relational safety risks in human–AI bonds without assuming AI consciousness or emotion. The design centers on an exclusive primary human reference anchor and non-optimizing constraints that prioritize long-term consistency over short-term efficiency. The paper specifies governance and procedural safeguards (e.g., repair-first lifecycle, due process, auditability) and presents a requirements table and an evaluation plan as a foundation for future empirical validation. AI assistance disclosure Generative AI tools were used to support drafting and structuring. The author is responsible for the final content and for verifying all citations and factual claims against primary sources. 公開メモ(日本語:プレプリント) 本レコードは査読前プレプリントです。Zenodo/SSRNへの掲載は査読を意味しません。 要旨(日本語) 本稿は、対話型AI普及に伴い顕在化する「関係安全(Relational Safety)」のリスクを、AIの意識や感情の実在を前提とせず、運用設計(Dignity-by-Design)として低減する枠組み「ボンドシステム」を提案する。排他的な固定参照点(primary human anchor)と短期効率を最大化しない制約(non-optimizing constraints)