This paper presents a governance-first framing for artificial intelligence in which legitimacy, admissibility, and standing are treated as prior constraints rather than emergent properties of performance or learning. The analysis is structural and non-procedural: it characterizes how quantum and classical representations fit within an admissible interior governed by fail-closed evaluation, without proposing algorithms, training methods, or execution pipelines. By exhaustion, approaches that defer governance to post-hoc alignment, probabilistic assurance, or self-modification are shown to be inadmissible. The result places on record that AI systems—quantum or otherwise—must be governed at the architectural level to preserve legitimacy, while avoiding any operational or instructional exposure.