Human in the Loop

An arrangement in which a person reviews an AI system's output before a decision is made, rather than leaving the system to decide alone. The Information Commissioner's Office (ICO) describes it as a process "in which a human considers the output of the AI model, as well as other information available to them, and then acts (makes a decision) based on this", and says decisions made using AI are "either fully automated, or with a ‘human in the loop’". The safeguard is only as strong as the review. A November 2026 Nuffield Family Justice Observatory report argues, of judges using Microsoft Copilot, that "behavioural research shows this safeguard is unreliable in practice", citing a study in which participants who met faulty AI output "accepted it uncritically 73.2% of the time". For businesses, a human reviewer is a real safeguard only if they have the time, information and authority to overrule the system.