Contemporary approaches to artificial intelligence governance emphasize ethics, policy, risk management, and post-hoc oversight. While these mechanisms may support transparency and accountability, they do not address the fundamental problem of computational authority: the ability to deterministically control whether an action is permitted to execute. This paper introduces binary governance, a scientifically grounded framework that defines governance as an execution-time control problem rather than a behavioral or interpretive one. Binary governance enforces authority as a strict precondition to execution, constraining state transitions through explicit, machine-resolvable permit-or-deny decisions. Prohibited actions are rendered non-executable rather than discouraged, detected, or corrected after the fact. The paper demonstrates that existing AI governance frameworks are structurally incapable of guaranteeing prevention because they operate outside the execution boundary where control is possible. By grounding governance in deterministic state-transition mechanics and physical execution constraints, binary governance enables enforceable authority, formal verification, auditability, and falsifiability across execution-capable computational systems. Although motivated by artificial intelligence, the framework applies broadly to software systems, embedded controllers, cyber-physical systems, and safety-critical infrastructure. The analysis is independent of model architecture, learning methods, or application domain, and focuses instead on the conditions required for governance to function as control rather than oversight. Elements of the execution-time authority and binary governance framework described in this work are the subject of a United States patent application filed in 2025, establishing priority for deterministic authority enforcement at the state-transition boundary.
Timothy M. Gough (Tue,) studied this question.