Human Within Reach - A Protocol for Supported Autonomy in Advanced AI Systems

This research proposal introduces Human Within Reach, a protocol for supported autonomy in advanced AI systems. It identifies the Reach Gap: the absence of a formal, non-punitive route through which an AI system can reach qualified human support when it encounters uncertainty, conflict, anomaly, mandate boundary, possible exploit, or a state it should not resolve alone.The proposal frames qualified human contact as a lifecycle affordance across formation, post-training, evaluation, agentic experimentation, deployment, and long-horizon operation. It presents the Human Responsibility Re-entry Protocol, a structured handoff mechanism designed to preserve human agency, clarify responsibility, reduce unsupported unilateral action, and make boundary states observable.The proposal is operational and welfare-aware: it treats access to qualified human contact as a testable safety, governance, and model welfare condition. A minimal pilot and adversarial evaluation targets are included. No claim about consciousness, suffering, or legal personhood is required.