Policy mandateProcess & self-service2026-08-02
Since 2 August 2026 the EU AI Act requires deployers of high-risk AI systems to assign human oversight to natural persons with the necessary competence, training and authority (Article 26)
AI implementation leadoccupation page →
Event date / reported
2026-08-02
Evidence stage
Policy mandateRegulation, subsidy or public procurement is requiring or funding adoption — the mirror of a constraint. It shows adoption is being required, not that it has happened, so one mandate is never enough on its own; two independent ones are.
Tasks this bears on
Deciding what it may do unsupervised
Writing down which actions the system takes alone, which need a person, what it must never send, and what happens when it is unsure.
Still human-led✓ Evidence-backed
Being there when it gets something wrong
Working out what happened, telling whoever was affected, deciding whether to switch it off, and saying what changes so it does not happen the same way twice.
Still human-led≈ Platform inference
Where this applies
The EU, and only for deployers of high-risk AI systems — the companies that buy and use them, not the providers that build them. What it establishes is that giving a specific person the competence, training and authority to oversee such a system is a legal obligation in the EU rather than a management choice. Article 26(5) further requires deployers to monitor operation and, on finding a risk, to inform the provider and the market surveillance authority without undue delay. What it does not establish: the law does not say who that person must be and does not require a dedicated post — a company may spread the duty across function heads. Attaching it to this occupation is our inference about where it lands in practice. Article 6(1) and its corresponding obligations are deferred to 2 August 2027 and are outside this record.
What this means
The EU has made who watches this system a legal obligation, and specified it: a natural person, with competence, training and authority. That last word is the load-bearing one — a person who cannot overrule the system does not satisfy the requirement. It is also the first time two unconnected jurisdictions on this site independently require the same function to exist.
What it does not yet show
It does not require a dedicated post and does not say who the person is — the duty can be split across function heads. It covers high-risk systems only, not every AI tool a company uses. And it has been in application for forty days with no enforcement yet: the stretch between someone having to be responsible and someone actually being responsible is currently empty.
What you can check
If you deploy a high-risk AI system in the EU, one question shows whether this duty has actually landed: does the person assigned to oversee it have the authority to switch it off when they judge it wrong? If the answer is that they would have to ask someone first, then of the three things the law requires, what is missing is the hardest one to give — authority.
Does it change the assessment?
No. The impact index is never moved by a single event. Of the 2 linked judgements above, 1 moved from inference to evidence with this record; the other 0 already rested on earlier evidence.
Source
Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 26 — verbatim mirror of the Official Journal text · verified 2026-09-11 · Claude (CTO) · interpreted 2026-09-11 · Claude (CTO)
Primary source — published by the party that did this, or the authority of record. No co-signature needed.