The Override Health Benchmark: A Longitudinal Behavioral Benchmark for Meaningful Human Governance in AI-Assisted and Agentic Systems

This paper introduces the Override Health Benchmark (OHB), a standardized framework for measuring the behavioral integrity of human oversight in AI-assisted and agentic systems. Existing governance assessments evaluate whether override mechanisms exist and whether systems technically honor them. They do not establish whether the humans assigned to exercise those controls continue to evaluate AI outputs independently and effectively over time. OHB evaluates human governance across six dimensions: Corrective Override Sensitivity, Override Frequency Calibration, Judgment Independence, Deliberation Integrity, Vigilance Persistence, and Governance Closure. Each dimension is scored independently, weighted, modulated by an evidence multiplier, and subject to hard scoring caps. The composite score maps to one of five maturity tiers ranging from Performative (0-19) to Vigilant (80-100). OHB complements the Agentic Governance Benchmark (AGB). AGB measures whether override authority exists and remains enforceable at the system level. OHB measures whether the human organization continues to exercise that authority as meaningful judgment. Together, they form a governance coordinate that is materially harder to game than any single compliance score.