top of page

The Gate Opens From Below

Nov 12
6 min read

Updated: 2 days ago


AI is handing control to the mass customer first. Higher up, the institution keeps it, and the wealthiest client holds the most while deciding the least.


Since June we have kept a weekly watch on artificial intelligence as it is deployed across banks, brokerages, wealth managers and family offices. Every item is graded for how well it is sourced, and nothing is written into a pattern until it has held across more than one cycle. Six months in, one asymmetry has held in every run. It is not about speed, though it looks like speed at first. It is about where the machine is placed, and in whose hand.


Where the gates opened

At the mass end of the market, the AI has been handed to the customer, and it has moved steadily toward the money. In June a British bank put a conversational assistant in front of twenty-one million accounts. In August a retail investing platform told analysts that a user could now authorise an AI agent once and let it trade within the scope of that grant. In September a European broker connected outside AI models to client accounts with a confirmation for every trade. In October Robinhood reported, by its own count, more than a hundred and fifty thousand accounts run by AI agents that can trade without individual confirmation if the user allows it. A robo-platform put AI on the advice itself, with no human adviser behind it. A consumer app announced "agentic finance" for the mainstream. And this month a Canadian platform launched an assistant that prepares transfers and trades and leaves only the final step to the client, with outside agents such as ChatGPT and Claude promised for next year.


The designs differ. Some ask for a confirmation at every step, some ask once, some let the customer switch the asking off. What they share is the hand on the switch. In every one, the customer holds it.


Where nothing arrived

At the top of the market the record is the opposite, and it is remarkably steady. Run after run, the watch has recorded the same absence: no dedicated agent for the family office or the ultra-wealthy client that advises, allocates or executes. What has arrived is real and substantial, but it faces the other way. Merrill's tools prepare the adviser for the meeting. JPMorgan's private bank has AI screen markets and positions overnight so that the banker can speak to the client in the morning. HSBC's private bank equips its relationship managers and keeps them as the client's interface. Deutsche Bank's private bank checks the source of a client's wealth with an agent and a person who decides. The family-office platforms read, consolidate, monitor and report; one describes its client tool, in its own words, as taking "no action on accounts." Where anything near the top moves toward action, it moves through the adviser, who runs the tool and answers for it.


Family offices are not indifferent to AI. In one survey of North American family offices, it was the top investment pick for the year ahead. But that is AI as an asset in the portfolio, not as a tool in the client's hand. The wealthy client's experience of artificial intelligence, so far, is of an institution that has become better informed about them.


AI lands where the decision already sits

The two records are not in contradiction. They are the same rule, applied to two different arrangements of control.


The mass customer already makes their own decisions. They open the app, choose the fund, press the button, on a small scale and often without thinking of it as control at all. When the machine arrives, it arrives where the decision is: in their hand. The wealthy client's arrangement is different. Their wealth is managed for them, often under a mandate that formally gives the institution discretion; the decision runs through an adviser, a relationship manager, an investment committee. When the machine arrives there, it also arrives where the decision is, which is on the institution's side of the table. They have far more, and far less of their hand on it.


This is why the opening reads as liberation at the bottom and as nothing at the top. The technology is the same. The arrangement it meets is not. An institution that holds the decision over billions has no structural reason to place an agent in the client's hand, and every reason to place it in the adviser's. We do not read that as hostility. The institution is a counterparty, not an opponent, and it is building what its own model of service requires. But the effect is plain: the wealthier the client, the further the new instruments sit from them.


The price of the open gate

The liberation at the bottom carries its own weight, and it should be named. When control passes to the customer, so does responsibility. One platform's disclaimer gives the user "full responsibility for every trade." A major American bank's updated agreement, effective this month, makes the customer "solely responsible" for what their AI agents do on the account. The mass customer gains the switch and the liability with it.


Nor is the mass market uniformly open. The large incumbent banks have moved the other way, keeping outside agents out of their customers' accounts altogether; in Canada, the big banks are reported to be doing the same. The gates have opened at the platforms whose business was already built on the customer deciding for themselves. That refines the rule rather than breaking it. It is the institution's own model of control that decides where the machine is allowed to land.


Not discretion, but direction

It would be easy to explain the quiet at the top by the discretion of private banking. We do not think that is the explanation. Since August 2026, the European AI Act has required that a person dealing with an AI system be told so, and a disclosure buried in terms and conditions does not count. Colorado's new law, due to take effect next year, will require notice where automated systems materially shape a consequential financial decision. These rules reach the machine that faces the client. They do not reach the machine that sits behind the adviser and prepares the meeting. At the top of the market, almost all of it sits there. Nothing obliges an institution to announce the intelligence that serves its own side of the table, because it is not the client who is speaking to it.


The silence at the top, then, is not a secret. It is a direction.


Does the line climb?

Our reading is that the opening began with the many and is moving upward. The evidence for the movement itself is still thin, and we hold it as a question rather than a finding. The first signs are in the middle: a client-facing avatar given to the affluent tier just below a private bank; advice delivered by AI to ordinary investors; the question, raised across the summer, of whether a household with a million under management still justifies a human adviser's time. If the line is climbing, it will show next in the mass-affluent band, then in whether outside agents reach affluent clients with the confirmation step intact, and finally in the first instrument at the private tier that hands the client the button rather than the adviser. We will be watching for each, and we will say so when one appears.


What it means for the principal

For the person we write for, the private decision-maker with substantial wealth, the asymmetry has an uncomfortable edge. The tools of independence are arriving fastest for those with least at stake, and slowest for those with most. Everything the institution is now building around the client makes it a more capable counterparty: better prepared, better informed, quicker to act on its own side. Nothing in that is improper. But it leaves the clients where they were, while the room around them grows more intelligent.


A Bespoke AI Confidante does not change that arrangement, and we would not claim it does. It does not take the decision from the institution, reach into any account, or press any button. What it offers is narrower. It is an intelligence formed from one person and turned toward them, so that while the institution grows more capable on its side of the table, the client is not left to meet it alone on theirs. The control the mass customer is being handed is control over transactions. The control that matters most at the top was never a transaction: it is the principal's own judgement about what their wealth is for. No institution can return that to them, because it was never theirs to hold. It is the clients’ to claim, and it is worth having someone on their side when they do.


Drawn from six months of structured, provenance-graded monitoring of AI deployment in finance, June to November 2026.


Founder & CEO of SMA Crown Confidential


Bespoke AI Intelligence for Private Decision-Makers


Comments


bottom of page