Abstract
AI is entering safety-critical work faster than most control systems were designed to absorb. In EHS and operational settings, AI increasingly classifies risk, prioritizes signals, routes work, recommends action, generates controls, monitors compliance, and shapes what people see, decide, and do. At the point where AI materially influences operational judgment or workflow execution, it stops being a tool and becomes a performer within the work system.
HAOP extends Human and Organizational Performance by recognizing three interacting performers: the human who adapts under real conditions, the AI that optimizes against data, signals, permissions, and constraints, and the organization that authors and governs the work system both perform inside. Its central claim is that safety-critical AI cannot be governed by model performance alone. It must be treated as part of a socio-technical system where human adaptation, AI optimization, and organizational signaling interact.
The paper defines the distinct failure signature of each performer, sets out Accountability by Control in place of blame, develops grounding as a condition of the work with anchors as its mechanism, and introduces the True Function Test: a diagnostic for whether an AI-enabled workflow produces the safety outcome it claims to pursue, or only a cleaner representation of it.
What Revision 4.1 adds
- Five hazard domains for AI-enabled work — a baseline operational register: physical interaction and machine agency, psychosocial conditions, human performance and verification, information and knowledge integrity, and work-system design and control allocation.
- The performer boundary — behavior as the genus (fire, humans, AI, and organizations all have it; none requires intent), performer as the species: behavior plus a delegated function in the work system.
- Eight operating principles and Accountability by Control: responsibility mapped to whoever holds control over a function, its design, and its operating conditions, before failure rather than after.
- Governance alignment — positioning HAOP alongside ISO 45001 and ISO 42001, with SIF classification as the downstream consequence layer.
- Part II: Applying HAOP — the work-design sequence the HAOP Workbook will operationalize: map the workflow, allocate functions, place verification gates, and assign control.