A governance framework that extends Human & Organizational Performance to the AI now making decisions inside safety-critical work.
In safety-critical work, AI now classifies risk, routes work, prioritizes signals, recommends action, and shapes what people see and decide. Most organizations still govern it as a data system, checked for privacy, accuracy, and uptime. Those controls are necessary and no longer sufficient.
Human & Organizational Performance corrected an older mistake: that the worker is the variable to control. HAOP extends that logic for a system with another performer in it. It recognizes three interacting performers whose failures look nothing alike.
Adapts under real operating conditions, and fails through overload, normalization, and silence.
AdaptsOptimizes the signal it is given, and fails through confident incompetence and unconstrained optimization.
OptimizesShapes both through incentives and metrics, and fails through drift, distorted signals, and normalized deviance.
SignalsA governance framework for AI-enabled safety-critical work. Introduces the three-performer model, the distinct failure signatures of each performer, and the True Function Test, a practical diagnostic for whether an AI-enabled workflow produces safety or merely a representation of it.
The first published articulation of HAOP for the safety profession. Why frameworks built to govern human performance do not, on their own, govern the AI now performing alongside it.
Scoping — run first
A tool for bounding a workflow before the diagnostics run: where it begins, where AI enters, which decisions it affects, and where consequences land.
A screen for whether the AI is still a tool or has crossed to performer-level by shaping priority, routing, approval, visibility, recommendation, or execution. When the call is ambiguous, it governs as a performer.
Diagnostics
A structured screen for whether a specific workflow has the basic conditions needed for responsible AI-enabled operation.
A diagnostic for whether a workflow delivers the safety outcome it claims or only the appearance of safety.
A test of whether “human in the loop” is real oversight, supported by time, competence, authority, data access, and the ability to intervene.
A check on whether the system’s optimized signal matches the stated objective, or whether the workflow rewards the wrong behavior.
A governance map of who controls each consequential action, signal, constraint, permission, metric, verification point, escalation path, and deployment decision.
For collaboration or speaking inquiries on governing AI in safety-critical work.