Human Sovereignty Under Machine-Speed AI R&D: Separating Human Acceptance, Technical Determinacy, Delegation, and Execution Authority

Calls for human oversight of advanced AI often assume that meaningful control requires humans to review each important action. That assumption becomes increasingly unstable when automated AI research and development operates faster than ordinary human decision cycles. This paper develops a human-sovereignty architecture for machine-speed AI R&D. The objective is neither continuous human micromanagement nor transfer of authority to AI systems. Instead, human and institutional authority remains the source of purpose, policy, delegation, responsibility, boundary change, and exceptional decision, while machine-time systems enforce previously authorized technical boundaries. The framework separates four states that are frequently collapsed: constituted candidate output, human or institutional acceptance, operation-specific technical determinacy, and execution authority. Human acceptance does not establish that a technical operation is admissible. Technical admissibility does not determine that an institution should perform the operation. Delegation does not imply authority to redefine the delegation boundary. AI recommendation does not become human judgment merely because an AI system possesses superior technical capability in a particular domain. The paper develops an explicit delegation grammar covering subject, object, operation, scope, duration, generation, transferability, and escalation conditions. It further distinguishes ex ante human sovereignty from machine-time enforcement, allowing routine activity inside previously authorized boundaries to proceed automatically while novel operations, policy changes, ambiguous evidence, and exceptional cases are escalated. The architecture addresses epistemic asymmetry in which AI systems may outperform individual human reviewers. Human authority does not depend on humans being technically superior to every machine system. Instead, AI systems may provide analysis, explanations, alternatives, uncertainty estimates, and evidence while human and institutional actors retain normative and organizational authority. The paper further examines accountability, typed human overrides, multi-party authorization, escalation packets, successor-system authority inheritance, delegation creep, approval laundering, technical laundering, automation theater, and adversarial testing of the human-machine authority boundary. The central principle is that human authority remains human, while enforcement of previously authorized boundaries may operate at machine speed. This publication is intentionally limited to public research-level abstractions. It does not disclose unpublished patent claim language, confidential claim charts, private source locators, provider-specific production parameters, non-public test vectors, or other confidential implementation details. Structural Paper Series - Paper 05 Final Publication Edition v2.1 Research Program on Deterministic Infrastructure and Human-Centered AI Coordination Transition Intelligence Institute, Switzerland Institutional Establishment in Preparation Foundational Research Signature: The Second Waters