Human, AI, and Organizational Performance (HAOP): A Safety Framework for the AI Era

Abstract

Human, AI, and Organizational Performance (HAOP) is a safety framework extending Human and Organizational Performance (HOP) to work systems in which AI performs consequential functions. The framework recognizes three performers in one work system – the human who adapts, the AI that optimizes within its representation and constraints, and the organization that contains and governs, authoring the conditions both perform inside – and provides the governance architecture that follows from seeing them together: characteristic failure and contribution signatures; a five-hazard register for AI-enabled work; five operating modes that classify how AI performs in the work; an Accountability by Control discipline that keeps responsibility attached to the roles holding relevant control; a work-design method organized by the industrial hygiene sequence ARECC; workflow design requirements for grounding through four anchor types, Verification Gates in human, automated, and hybrid form, Anchor Access, the protected ability to pause, and the Designed AI Reassessment Pause; a bidirectional hazard-communication requirement carried by the AI Deployment Safety Data Sheet; candidate leading measures for Verification Overrun as a hazard condition; and the True Function Test, an initial diagnostic that can be applied now. Companion deposits: Verification Overrun: The Safety Hazard Inside AI Oversight (concept DOI: 10.5281/zenodo.20874288; resolves to the latest version); The Collapse of Correction: Closed Informational Loops in Safety-Critical Work, v1.0 (DOI: 10.5281/zenodo.20856122); From Human-in-the-Loop to Human-in-the-Design, v1.1 (DOI: 10.5281/zenodo.21633748), supplying the Verification Requirement and Capacity apparatus and the three design obligations. The concept DOI https://doi.org/10.5281/zenodo.21631544 resolves to the latest version; The HAOP Method: ARECC Applied to AI-Enabled Work Systems, v1.0 (DOI: 10.5281/zenodo.21868086), supplying the framework's application method.The concept DOI https://doi.org/10.5281/zenodo.21868085 resolves to the latest version Changes in Version 6.0 (August 23, 2026) appear in the document itself before the Table of Contents. Version 6.0.1 (August 23, 2026) includes an updated AI Assistance Disclosure Statement above References.