Authority Boundary Test (ABT) v1.0: A Pre-Deployment Conformance Standard

Authority Boundary Test (ABT) is a pre-deployment diagnostic standard designed to detect authority leakage in systems, policies, workflows, and claims involving AI, automation, or machine-assisted verification. ABT evaluates one structural condition only: whether evidence, recommendation, or verification has been allowed to function as authority in consequential decision-making or execution. ABT is a boundary diagnostic, not a general framework for AI safety, fairness, ethics, legality, or model performance. A system may be accurate, robust, or compliant and still fail ABT if evidence functions as live authority. Version 1.2 preserves the core rule and pass/fail structure of prior versions while clarifying application to chained, tool-using, and agentic systems. This revision strengthens authority mapping across multi-step execution pathways, re-evaluation triggers after material system changes, refusal and escalation requirements, and accountability controls for downstream automated action. ABT is intended for use by organizations, auditors, operators, policymakers, and researchers evaluating whether authority remains explicit, attributable, revocable, and externally accountable where consequential action becomes possible. Core Rule: Verification is evidentiary. Governance is authoritative. Verification may inform decisions. It must never exercise them.