Who is accountable?

When an AI prepares a letter, an analysis or a decision, accountability remains human. But it still has to be assigned. Organisations that succeed write down who can delegate what, who approves before sending and who answers for it in case of error.

A simple three-level rule

• What the agent does on its own: search, summarise, prepare drafts.
• What requires approval: sending, publishing, paying, deleting, changing a shared calendar.
• What remains human: decisions that affect a person’s rights, such as a hiring, a sanction or a diagnosis.

The UNESCO Recommendation on the Ethics of Artificial Intelligence, adopted by its member states in 2021, sets out the same principle: ultimate responsibility must always be attributable to a natural or legal person.