SMA Crown Confidential Notebook – Note: EU AI Act

Occasion: The EU AI Act's high-risk obligations became binding on 2 August 2026, making human oversight a statutory requirement for automated decisions affecting access to financial services — with "human oversight" defined (Article 14) as the capability to intervene and override, not per-decision sign-off.

A definition became binding across the sector this month. The EU's AI Act now requires human oversight of the automated decisions that reach into a person's financial life — credit, fraud, access to financial services — and it defines that oversight with some care: the ability to intervene and to override, rather than a signature on each individual decision. A capacity to step in, not a duty to sign off. The distinction will occupy the sector for some time, and it should; it is the difference between a human who can act and a human who must.

We would only note where all of this attention is pointed. It is oversight of the decision — the move, the transaction, what to do with the holdings. The entire apparatus now being built and audited across the industry, the thresholds and intervention rights and the human kept near the switch, is arranged around that layer: the place where something is done with the assets, and where the question is whether a person remained close enough to the doing.

It is a serious layer, and worth governing. It is not the one we work in.

There is a question beneath every decision about what to do with wealth, and it is a different kind of question. Not what to do with what you hold, but who to be with it — what the wealth is for, and the kind of steward you intend to be in holding it. That question is not delegated to a machine, and so it is not one any oversight regime is written to govern. There is no threshold to set on it, no override to keep in reserve, no switch a human must stay near. It never left the person's hands, because it is not the sort of thing that can be automated in the first place. Deciding, at this depth, is inseparable from holding.

This is the line we keep, and the season's news throws it into relief. The measures now confirming that deployed AI does far less on its own than its marketing suggests — that the working mode remains collaborative, the human retained — are all describing the same layer: the decision, and how much of it a machine may take. That debate matters, and we do not stand outside it lightly. We simply begin elsewhere. Oversight keeps a human above the transaction, able to intervene, entitled to override. We keep the person underneath it — with the question that was never the machine's to hold, and never needed watching, because it was never handed away.

Seraphina Michelina Evelyna Bogomolova-Huotelin

Founder & CEO of SMA Crown Confidential

Digital Confidantes: Bespoke AI Intelligence for Private Decision-Makers