Art. 26: what deployers of high-risk AI actually owe
Last updated: · Leutrim Miftaraj
Art. 26 of the AI Act obliges deployers, among other things, to use systems as intended, to ensure human oversight by competent persons and to cooperate on record-keeping. For agentic systems this means: the agent's limits of authority are part of intended use — and must be provable.
From document to enforcement
A policy saying «the agent may spend at most X» meets the duty on paper; a mandate meets it in operation: the limit is checked on every action, breaches are rejected and logged — oversight as a runtime property.
Competent oversight person
The escalation inbox operationalises the role: defined thresholds route decisions to named people, with context, deadline and a documented outcome — who oversees is provable, not merely designated.
Working with providers
Providers deliver systems with oversight interfaces (Art. 14); deployers use them. The mandate is the interface that connects both duties — and the export is the proof towards market surveillance.
Key takeaway
Art. 26 asks for oversight that is lived. Mandates make it a system property with proof.
Try it yourself
Issue your first agent mandate and check your first decision with its receipt — free, in a few minutes.