Human control means people retain meaningful authority over consequential AI-assisted actions. It is more than putting a person somewhere in the process.
Approval: identify which actions need a person to authorize them. Stop: provide a way to halt an action or workflow. Override: let an authorized person correct or replace the AI's proposed action. Rollback: define how to recover when an action was wrong.
A low-risk suggestion does not need the same process as a financial, medical, employment, safety or infrastructure decision. The control burden should increase with the possible consequence and irreversibility.
Human owner, permissions, limits, approval points, logging, escalation, stop conditions and recovery path.