The Dual-Track AI Governance Framework: Capability Control, Authorization, and Human Responsibility

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

WP-AI-8.3 develops a dual-track framework for AI safety and governance by connecting two dimensions that are often treated separately: technical capability control and institutional human responsibility. The first track governs what an AI system can do, where it may operate, what authority it may exercise, and when that authority must be reduced, suspended, or reviewed. The second track identifies responsibility across development, release, regulatory admission, institutional configuration, and operation. The framework then cross-maps these two tracks so that every significant capability boundary corresponds to an identifiable authority, authorization decision, evidence path, and responsibility point. The paper introduces several core governance principles: capability control without responsibility produces unmanaged risk; responsibility without real control produces empty governance; authorization should have a lifecycle rather than remain permanent; and material changes in capability, environment, configuration, or action authority should trigger renewed governance review. WP-AI-8.3 builds directly on WP-AI-8.1, which developed the engineering circuit-breaker principle, and WP-AI-8.2, which examined AI responsibility escape. It serves as the framework-centered paper of the current stage of the WP-AI-8 series, Foundations of AI Safety and Governance.