Who is responsible when AI makes a mistake?
Responsibility is a legal and organisational question, not a property of the model.
What is it?
When AI makes a mistake, the model is not responsible; people and organisations are: the developer, the provider, the company using it or the person who made the decision. Responsibility has to be settled clearly before a system goes into use.
An everyday example
A hospital uses software to triage emergencies. If it misses a serious condition, the question of liability concerns the hospital, the manufacturer and how its use was supervised.
Why does it matter?
Without clear responsibility, people who are harmed do not know whom to turn to, and organisations have no incentive to be careful.
A common misconception
That a notice saying “AI can make mistakes” shifts responsibility to the user. A warning does not replace due care.
In more detail
Can we blame the model?
A model is no substitute for the person, company or institution that has the obligations.
Who can be held responsible?
That depends on the legal regime and the facts: the provider, the deployer, the user or another responsible party.
What does meaningful oversight look like?
Clear division of responsibilities, documented decisions, testing, the ability to stop the system and a way to complain.
Is a notice saying “AI can make mistakes” enough?
Not as a general solution. The obligations depend on the risks, the context and the law that applies.
Remember: Responsibility has to be settled before use, not only after a mistake.
Reliable sources
Last reviewed 1 November 2026. An explainer, not professional, legal or medical advice. For rules and figures that change, check the official version.