In 1983, Stanislav Petrov was covering someone’s shift at a Soviet nuclear early-warning bunker outside Moscow. Suddenly, a siren sounded and his screen lit up with a single word: launch. According to the system, the US had fired five missiles, heading their way.

Following protocol, he should have reported it up the chain – a call that could have triggered retaliation and potential nuclear war. But he paused. The alert felt suspicious, and the other satellites hadn’t spotted anything. He called army headquarters and reported it as a system malfunction. Twenty-three minutes later, no missiles had landed. His judgment was right.

“They were lucky it was me on shift that night,” he said. It was a flaw in the system, caught by the one thing the system lacked. Technologists now have a name for what saved the world: human-in-the-loop.

What is human-in-the-loop?  

Human-in-the-loop – or HITL – means keeping a person involved in an AI process, rather than letting the system run unchecked. The human might review the data, test the output, approve a recommendation or step in when the machine gets it wrong.

Done properly, it is more than a rubber stamp. The person in the loop needs enough understanding, authority and independence to challenge the system, not just click ‘approve’.

AI can process data and detect patterns at speed. But it cannot weigh context, exercise discretion or spot every error that would be obvious to someone with real-world experience. 

Where do we see it in practice?  

In healthcare, AI tools screen medical images for signs of cancer and flag deteriorating patients in intensive care. In finance, they inform loan approvals. On social media platforms, they moderate content at a speed no human team could match. In some criminal justice systems, they have been used to inform sentencing decisions. 

Imagine any of these without a human in the loop.

For example, when journalists at ProPublica investigated the COMPAS system used in US courts in 2016, they found the software was significantly more likely to incorrectly flag Black defendants as future criminals. Left unmoderated by a human, it could have led to a severe injustice in the legal system.

A human-in-the-loop framework doesn’t eliminate all risks. But it ensures that a person – not an algorithm – bears the weight, and the accountability, of a decision.

Why does it matter for businesses?  

Most business decisions don’t carry life-or-death stakes. But the same logic applies wherever AI shapes outcomes that affect people. The principle is now being written into law: the European Union’s AI Act, which came into force in 2024, makes human oversight a requirement for high-risk AI systems.