The Open-Box Fallacy: Why AI Deployment Needs a Calibrated Verification Regime

Abstract

AI deployment in sensitive domains is often treated as unsafe to authorize until model internals can be explained. We argue that the gate should instead be calibrated verification: domain-scoped, independently checkable, monitored, accountable, contestable, and revocable. We propose Verification Coverage, a six-component reportable standard with a minimum-composition rule, intended to sit beside capability scores in model cards, leaderboards, and regulatory disclosures.

Download Paper