Human override rate
A measurable indicator of whether a human review checkpoint is substantive: the frequency at which a human rejects or modifies an agent's proposed action. A rate near zero over a long period, especially paired with very fast review times, is a signal of automation bias rather than evidence of genuine oversight, and is one of the specific practices the Version 1.5 update added for auditing the effectiveness of human oversight.
Defined in 2 GAGE programs, which carry 4 distinct definitions of it. The wording above is taught in AI Governance: Applied Mastery.
How each discipline defines it
The same term does different work depending on who is using it. These are the definitions as each program teaches them, unedited.
A measurable indicator of whether a human review checkpoint is substantive: the frequency at which a human rejects or modifies an agent's proposed action. A rate near zero over a long period, especially paired with very fast review times, is a signal of automation bias rather than evidence of genuine oversight, and is one of the specific practices the Version 1.5 update added for auditing the effectiveness of human oversight.
The frequency at which humans reject or modify agent actions during review. A low rate can indicate genuine agent quality or, per the MGF's own warning, rubber-stamping behaviour where a human checkpoint has stopped functioning as real review.
"The frequency at which humans reject or modify agent actions" (MGF p.30). A low or falling rate may signal rubber-stamping; the metric requires a baseline and ongoing tracking to be meaningful.
"the frequency at which humans reject or modify agent actions," a metric the framework names as a signal that "a low rate may signal rubber-stamping behaviours" (MGF v1.5, p.30).
Where it is taught
The exact lessons this term appears in. The first 7 topics of every program are free with a free account.
- Automation Bias · Meaningful Human Accountability, Agentic AI Governance: Applied Mastery
- Dossier Workshop D: Accountability Map · Meaningful Human Accountability, Agentic AI Governance: Applied Mastery
- Watching an Agent Work · Rollout, Monitoring, and Incident Response, Agentic AI Governance: Applied Mastery
- The Singapore stack: voluntary on paper, adopted in practice · The World's Rulebooks, AI Governance: Applied Mastery
Terms it appears with
Not an alphabetical neighbourhood: these are the terms taught in the same lessons, ranked by how often they appear together.