Override rate — the oversight vital sign

The short answer. Override rate is the share of AI recommendations clinicians decline, change, or act against. It is the vital sign of human oversight: near zero suggests rubber-stamping or alert fatigue in disguise; very high suggests a model clinicians have stopped trusting. Both extremes are findings about the system — the useful move is watching the trend, not judging the number.

If human in the loop is the control, override rate is how you check its pulse.

A rate near zero has two readings, one fine and one not: either the model is genuinely excellent, or the humans have stopped looking — automation bias, alert fatigue, or a workflow where disagreeing is expensive. A very high rate also has two readings: a model performing badly, or one performing fine on a population it was never meant for. The metric cannot distinguish these alone; its job is to raise the question, which is why it belongs on the monitoring dashboard next to the performance metrics rather than in an annual report.

The governance move is to trend it per model, ask for an explanation at the extremes, and record the answer — one more line in the monitoring evidence a surveyor can date.

Where do you actually stand? The free RUAIH readiness score maps your organisation against the five focus areas in about eight minutes, and the published crosswalk shows how each control lands across RUAIH, CHAI and the NIST AI RMF.

Override rate — the oversight vital sign

Keep reading