Human, AI, and Organizational Performance (HAOP) is a safety framework for AI-enabled work systems where human judgment, AI optimization, and organizational control jointly shape operational outcomes. This white paper extends Human and Organizational Performance (HOP) for environments where AI systems do more than store, summarize, or retrieve information. When AI classifies risk, routes work, prioritizes signals, recommends controls, approves actions, or shapes what human reviewers see first, it becomes a performing element within the work system. HAOP identifies three interacting performers: the human performer, who adapts under real operating conditions; the AI performer, which optimizes based on data, signals, permissions, constraints, and architecture; and the organizational performer, which shapes both through governance, incentives, resources, metrics, authority, procurement, and tolerated tradeoffs. The paper defines distinct failure signatures for each performer, introduces accountability-by-control, and argues that safety-critical AI cannot be governed by model performance alone. It must be governed as part of a socio-technical system where grounding, verification, pause authority, and accountability are designed before AI-shaped outputs become consequential. The white paper also introduces the True Function Test, an initial HAOP diagnostic for evaluating whether an AI-enabled workflow produces the safety outcome it claims to pursue or merely produces a cleaner representation of safety. Rev 3 (2026-07-02): Terminology standardized to "illusory function"; title page matter and version label corrected; Gibson v Maritime New Zealand 2026 NZHC 813 citation verified; grammar and typo corrections.
Jaina Ko (Thu,) studied this question.