Subject areaRules, risk and compliance
Also known ashuman supervisionhuman control
Human oversight is the control that competent people exercise over an artificial intelligence system while it is in use, to prevent or minimise risks to health, safety and fundamental rights.
Article 14 of the EU AI Act requires it for high-risk AI systems and asks the provider to design them so that it is possible.
Oversight that exists only on paper is useless. The regulation asks that whoever controls understands the capabilities and limits of the system, stays aware of the tendency to over-rely on its output (known as automation bias, the habit of taking a machine's answer as correct), can interpret the result, decide not to follow it and interrupt the operation with a stop command.
On the other side, Article 26 gives the deployer the task of entrusting oversight to people with adequate competence, training and authority. The theme runs through AI governance: the final responsibility for a decision stays with a human being and an identifiable organisation.
Source: Regolamento (UE) 2024/1689 (AI Act), articolo 14 e articolo 26, testo su EUR-Lex.
An example
A credit department uses software that suggests whether to grant a loan. The assigned officer knows the limits of the tool, can challenge its output and can stop the procedure: the final decision is hers, not the software's.
broader entriesthis entry and narrowertwo levels down← up · → down
Enrico Giubertoni. Licensed under CC BY-NC 4.0: you may quote, translate and adapt this entry with the author’s name and a link, stating any changes. For commercial use, write to info@enricogiubertoni.com
Spotted an inaccuracy or something that needs updating? Write to info@enricogiubertoni.com and help make this glossary a better place.
What are your business challenges?
Tell us about your priorities and the objectives you want to reach.
You will receive a free, targeted answer within one working day.