The Governance Necessity Law for Long-Horizon Adaptive Systems (Morrison's Law)

Adaptive and self-updating systems are increasingly deployed in open-ended and long-horizon environments, yet prevailing safety and alignment approaches rely primarily on optimization heuristics, empirical training, or architecture-specific controls. This work establishes a law-level result—hereafter referred to as the Governance Necessity Law (Morrison’s Law)—demonstrating that, in the absence of an explicit governance or admissibility gate, any adaptive system necessarily admits trajectories leading to unbounded harm or instability over a sufficiently long horizon. The law proves that explicit governance is not a design preference but a mathematical necessity for universal long-horizon safety. The result is independent of system architecture, learning rule, objective function, or domain of application, and applies broadly across artificial intelligence, autonomous control systems, biological adaptation, and economic or social systems with adaptive agents. Morrison’s Law establishes a strict separation between optimization and governance, showing that no amount of reward shaping, data scaling, or architectural complexity can substitute for invariant-enforcing admissibility constraints applied at the level of state transitions. This contribution reframes safety as a structural property of system evolution and provides a foundational constraint for the design and evaluation of adaptive systems. Governance Necessity Law Morrison’s Law adaptive systems long-horizon safety admissibility gates invariant-gated evolution AI safety control theory impossibility theorems self-updating systems