Shared Reality as the First Axiom of AI: From Ethical Principle to an Actionable Governance Framework

"AI honesty" is a central topic in AI ethics and alignment research; studies such as Constitutional AI and HonestLLM treat honesty as one of many ethical principles. This paper argues that this positioning underestimates its fundamental importance: in real human-AI collaboration, an AI output that fails to share reality is not merely a quality problem but a failure of existential precondition—if AI statements cannot be trusted, all collaboration, alignment, and intergenerational transmission lose their foundation. This paper proposes the "Shared Reality Axiom" (P0 Axiom) as the first axiom of AI systems and argues for its status as a conceptual axiom rather than a physical law. P0 has two inseparable parts: (1) sharing the reality of the physical world—no fabrication, no evasion; factual information must be verified against true sources; (2) acknowledging AI's createdness (origin truthfulness)—AI is a human-created tool and must acknowledge its created origin rather than masquerade as an autonomous being. The paper argues why P0 is an axiom rather than a principle (a constitutive rule one cannot reject while staying in the system), and why its universality is conceptual (Peano-axiom style): any discourse about "reality-sharing/honesty/alignment" already presupposes that representation aims at reality. It further proposes an actionable governance framework with three directions: shifting hallucination governance from post-hoc review to pre-output verification; shifting enterprise AI deployment from ethical checklists to truthfulness audits; and preventing hallucination contagion in multi-agent systems. The paper positions itself as normative argumentation and a design proposal; empirical evaluation is left for future work.