Meaningful control
A reviewer can detect problems, intervene and access a safe alternative.
05/ 09
Knowledge area
Clinical Safety & Human Oversight
What must people understand, review, override and escalate?
Design human involvement as a safety control with appropriate competence, time, information and authority—not as a disclaimer placed after an automated output.
Operating principles
What good
looks like.
Competence
Users understand intended use, limitations and signs of unreliable behaviour.
Proportionate explanation
Information supports the decision a person actually needs to make.
Contestability
Affected people can question outcomes and reach an accountable human.
Put it into practice
Make it
operational.
Map the decision pathway
Identify where AI influences prioritisation, diagnosis, treatment, access or communication.
Engineer oversight
Define review points, confidence cues, override, escalation and downtime procedures.
Validate with users
Test usability, automation bias, alert fatigue and realistic performance.
Maintain competence
Provide role-based training, supervision, feedback and periodic re-authorisation.
Minimum evidence set
Make the decision
reviewable.
01Clinical safety case
02Human-factors and usability evaluation
03Instructions, limitations and training records
04Override, escalation and downtime procedures
Move from knowledge to practice
Design governance teams can operate.
Connect accountability, evidence and oversight to the way AI is actually delivered and used.
Start a conversation ↗