Frameworks
Most conversations about AI trust stay at the level of opinion. Executives say users trust the system, users say they don’t, and a dashboard shows usage without showing why. A framework settles the argument by specifying what to measure, how to measure it, and what the results mean. These frameworks are built so that the output can be used in a decision, whether that is a deployment, a governance review or a dispute over a single case.
Each framework turns a behavioral question into something that can be measured, compared and acted on. They underlie every Behavieural engagement, and each rests on published behavioral science.
-
Oversight That Works: Five Conditions for Sustained Human Oversight of AI
11 October 2026
Putting a person in the loop is the start of oversight, not proof of it.
Nearly every AI governance regime asks for a human in the loop, yet few say what must be true for that person to catch the errors that matter. A reviewer who is present but rushed, under-informed, over-trusting, unable to act, or never told how earlier decisions turned out provides the appearance of oversight and little of its substance. Oversight That Works sets out a framework for the conditions that determine whether human review does its job.
The framework proposes that a reviewer can sustain effective oversight only when five conditions hold: attention to the case, understanding of what the system is doing and where it fails, calibrated confidence in its output, agency to act on a concern without penalty, and learning from the outcomes of past decisions. These conditions operate within an institutional envelope of workload, authority, and feedback that the reviewer does not control. It also advances a weakest-link hypothesis, under which oversight capacity is set by the weakest condition and not by the average. The framework keeps AI performance, oversight performance, and decision outcome distinct, so that a good outcome is never mistaken for good oversight.
The framework draws on research on human-automation interaction, vigilance, trust, and organizational behavior, and it maps each condition to the EU AI Act, the NIST AI Risk Management Framework, and ISO/IEC 42001. It is written for assurance providers, risk and compliance leaders, standards drafters, and the organizations that deploy AI in regulated settings, and it shows where to look, what evidence to ask for, and where to intervene when a condition is weak.
-
AI Trust Axis
1 June 2026
A framework for measuring how people come to trust and rely on AI.
The AI Trust Axis explains why technically sound AI systems still stall, get overridden or get quietly avoided. Instead of treating trust as a single number, it breaks the human response to an AI system into five behavioral dimensions, each tied to a distinct and recognizable failure: automation bias, algorithm aversion, escalation avoidance, shadow workflows, and performative human review. Because each dimension points to a specific mechanism, a weak result tells an organization what to change, not only that something is wrong.
The framework is operationalized through the AI Trust Axis Scale, a 25-item survey instrument. It produces two scores. The Trust Integrity Score assesses a deployed system, showing how well reliance holds up in real use. The Trust Readiness Score assesses a system before deployment, showing how prepared the people, workflows and governance around it are. The scoring method is proprietary.
The AI Trust Axis is the analytical basis for the BEAR, STAR, B-GRIT, and BOLD engagements, and for the research reported in Behavieural’s white papers and survey reports. The scale is scientifically-validated and published at Australasian Marketing Journal.