Who Keeps Watch
Updated: 4 days ago

AI on the firm’s side is watched, logged and approved. On your side, the watch should be yours to keep.
In September 2026, Ritholtz Wealth Management, one of the better-known American advisory firms, chose Hadrius, a compliance-technology company, to oversee its firmwide rollout of Claude. Hadrius monitors how the firm’s employees use the AI, reads their activity through the model’s compliance interface and archives every interaction under SEC and FINRA record-keeping rules, the way firms have long archived email. The oversight was bought from a third company, and it came with the rollout, not after it.
The same month, Accenture launched Trusted Wealth Ops, agentic workflows for wealth firms built on Salesforce and Claude. Its announcement gives the limits as much care as the agents: the work is “done with guardrails in place like permissions, approved data sources, human approvals, exception routing, and audit logs,” while “advisors and compliance teams remain in control of consequential decisions.”
The regulators followed. Reporting on the SEC’s 2027 examination priorities in October, WealthManagement.com wrote that the agency “is intensifying its efforts to understand how the technologies influence outcomes,” and a compliance head quoted in the report flagged employees who use personal AI accounts for business as a record-keeping concern. In the United Kingdom, the FCA’s observations on AI, as summarised for asset managers, expect a firm to explain “what a tool does, why it is used, who owns it and how its risks are controlled.” Governance should come before deployment, and to “launch first and attempt to bolt on governance afterwards” is named as poor practice. Senior managers remain accountable, and outsourcing “does not transfer responsibility.”
As such, it is expected, and it is right. An institution that puts AI into its client work should know what the AI does, keep a record of it and answer for it.
Now take it from the chair of the principal — your chair. Ask whom all this governance protects. The monitoring serves the firm’s compliance team. The archive is kept for the examiner. The guardrails keep consequential decisions with the adviser and the firm. You appear in it too, but as its subject: in Accenture’s description, agents “continuously monitor client portfolios” for signals such as funding gaps or lifecycle events. The governance watches over the AI that watches you. None of it is built to watch over anything on your behalf.
Then look at what governance exists on your side. In the previous piece in this series, a bank opened client accounts to general AI assistants with a single sentence of governance: the outputs are not ours, not advice, and at your own risk. The regulators are concerned about employees who bring personal AI accounts to work. No one asks about the AI you use at home to think through your affairs: what it knows, what it is told, what it keeps, or where it stops.
Yet the firms have already written the framework, and it translates with almost no change. Permissions: what the AI beside you may do, and what it may never do. Approved data sources: the material it works from, chosen by you rather than whatever it happens to be handed. Human approvals: the points at which every judgement returns to you. Exception routing: where it must stop and say it does not know. Audit logs: a record of what it told you and why you decided as you did. Add the FCA’s four questions, what it does, why you use it, who owns it and how its risks are controlled, and you have the governance of one person’s intelligence layer. Answered for an institution, that list is compliance. Answered for one person, it is formation. And the poor practice the FCA names applies to you as much as to any firm: launching first and governing later.
From your chair, the principal’s, the questions are these. Could you answer the four questions about the AI you already use? Who holds the record of what it has told you? Where has it been told to stop? And if it gave you poor counsel tomorrow, who would notice?
The first answer is practical. Before any AI touches your affairs, answer those four questions in writing, for yourself. If you cannot, it is running on its defaults, and its governance belongs to no one.
The deeper answer is a Bespoke AI Confidante governed by you: formed around your intentions, working from material you approve, stopping where your judgement begins, and keeping a record of your reasoning that belongs to you. Governance on your side is not a compliance department. It is authorship, done once with care and tended over time.
That is what SMA Crown Confidential forms. Not another adviser at the table, but an intelligence layer on your side of it: a Bespoke AI Confidante running on the model of your choice and formed around your own intentions, so that your judgement stays independent of every firm across the table. Over this series we have watched that table from your side. The firm keeps the minutes, holds the map, sets the dial and charts the course; a bank has opened the seat beside you; and now the firm keeps watch over its own AI. Every piece has asked the same question from a different angle. The last asks it plainly: who keeps watch on your side?
Seraphina Michelina Evelyna Bogomolova-Huotelin (nee von Prittwitz)
Founder & CEO of SMA Crown Confidential
Bespoke AI Intelligence for Private Decision-Makers
.jpeg)



Comments