Guardrail 05 of 10 · The Quantrim Guardrails Framework

A person can always step in.

Digital staff augment your team, they do not replace judgement. Human in Control means anything that touches a customer, a payment, or a decision that matters has a review point built in, and a person who can override the system at any time.

A pilot's hands on the controls of a small aircraft cockpit at sunset
In practice

What human oversight actually looks like

Review points on the things that matter

Anything touching a customer, a payment, or a consequential decision has a human review point designed in from the start.

Judgement calls escalate

Judgement calls escalate to a person, they do not get quietly automated away because it would be more convenient.

A real off switch

Every system has an off switch, and the named owner knows how and when to use it.

Trained on when to check, not just how to operate

Staff are trained on when to trust the system and when to check it, not only on which buttons to press.

Why it matters

The last 10% of judgement is also usually the highest-stakes 10%

Most volume can run autonomously without issue. It is the smaller share of judgement calls, the ones with real consequences for a customer, that need a person in the loop, and this guardrail makes sure that share is identified in advance, not discovered after something goes wrong.

Grounded in

  • Australia’s Voluntary AI Safety Standard, guardrail 5 (human control and intervention)
  • EU AI Act, Article 14 (human oversight, where applicable)

Read the full Guardrails Framework →

FAQ

Frequently asked questions

Doesn’t a human review point slow everything down?

Only the parts that need it. Most volume runs autonomously. It is the smaller share of genuine judgement calls that stay human.

What if the system and the human disagree?

The human wins. That is the entire point of the guardrail.

Find out where your governance actually stands. Governance is one of six things the free audit scores.

Get the free audit