Human Oversight Must Have the Power to Intervene

Institutions increasingly reassure the public that a human remains in the loop. The phrase sounds protective. It suggests that a teacher, pastor, officer, clinician, manager or board member will prevent an AI system from making an unacceptable decision.

But a person can appear in a workflow without exercising meaningful judgment. The reviewer may see only a recommendation rather than its evidence. The interface may encourage rapid acceptance. The workload may make careful scrutiny impossible. The reviewer may lack subject knowledge, technical understanding or authority to stop the system. A nominal signature can legitimise automation without governing it.

The decisive question is therefore not whether a human is present. It is whether that person can understand, challenge, interrupt and repair. Human oversight must have the power to intervene.

On 15 October, the Financial Stability Board published its final Sound Practices for Responsible Adoption of Artificial Intelligence. The report presents twelve practices spanning strategic direction, accountability, documentation, materiality, technology selection, data governance, explainability, performance, human oversight, cyber risk and third-party dependence. Financial Stability Board

The FSB describes effective oversight as meaningful rather than nominal: people need sufficient ability, authority and incentive to intervene. For agentic AI, it recommends risk-based approval triggers, controls between discrete steps, tested mechanisms to pause or remove access, escalation, fallback and explicit accountability. It also warns that agents do not reason like humans…