Foundational Fictions in AI Governance: Why Useful Untruths Require Structural Honesty

Foundational Fictions in AI Governance: Why Useful Untruths Require Structural Honesty This paper introduces the Fiction Pathology Test, a diagnostic framework for identifying when institutional fictions about AI systems become pathological rather than conventionally useful. Institutions routinely treat large language models as decision-makers ("AI decided") while simultaneously denying AI possesses decision-making capacity—a contradiction that enables responsibility deflection across corporate structures. Drawing on philosophical analysis of fictions (Vaihinger, Searle) and case studies from financial services, healthcare, and autonomous systems, the paper distinguishes:1. Useful fictions (corporate personhood conventions enabling coordination)2. Pathological fictions (AI agency pretense eroding accountability) The test provides regulators, lawyers, and governance professionals an operational mechanism to identify deflection-enabling contradictions, with concrete policy recommendations for structural honesty requirements in AI deployment. SSRN: 6015854