Autonomous AI agents are operating at scale across enterprise, government, and critical infrastructure systems. These agents make hundreds or thousands of consequential decisions per hour without per-decision human review. Current governance frameworks, including SOC 2, ISO 42001, NIST AI RMF, and the EU AI Act, define compliance requirements but do not measure whether those requirements are enforced at runtime on autonomous systems operating at machine speed. This paper introduces the Agentic Governance Benchmark (AGB), a standardized scoring framework that evaluates an organization's ability to govern autonomous AI agents in real time. The AGB measures six dimensions of governance readiness: policy determinism, enforcement latency, receipt provenance, scope containment, jurisdictional enforcement, and override integrity. Each dimension is scored independently on a 0-100 scale and weighted to produce a composite score that maps to one of five maturity tiers ranging from Ungoverned (0-14) to Sovereign (90-100). The benchmark is vendor-agnostic and designed to be administered against any agentic AI system regardless of model provider, deployment topology, or regulatory jurisdiction. A reference implementation based on the ExecLayer enforcement architecture demonstrates that Sovereign-tier governance is achievable with current technology. The AGB fills a critical measurement gap in the governance landscape by providing the first standardized method for quantifying runtime enforcement capability in autonomous AI systems.
James D. Benton (Mon,) studied this question.