.jpeg)
SMA Crown Confidential Notebook – Note: Ultimate Accountability
Occasion: The Financial Stability Board delivered its sound practices on agentic AI to the G20 on 12 October (non-binding; the specific recommendations — human-in-command, ultimate accountability on a named institution or person — are as reported from the consultation the final report finalises, not quoted from the Chair’s letter). Alongside it, the Monetary Authority of Singapore issued its AI Risk Management Guidelines on 7 October: “engaging a third-party AI provider does not reduce a financial institution’s accountability”; the agent-specific human-in-the-loop rule is deferred to a 2027 consultation. Both regulators’ placing of accountability on the institution is taken at face value.
Position occasioned: on the client’s side of the table. A chain-level lead, taken because the regulatory pair dominated a deployment-light week, and grounded in the client by the closing turn. However well governed, the institution remains a counterparty; the Bespoke AI Confidante has no institution behind it and no interest of its own.
This week the Financial Stability Board delivered its sound practices on agentic AI to the G20, and the Monetary Authority of Singapore issued guidelines of its own. They are different instruments, one a standard-setter’s recommendations, the other a supervisor’s rules, but they share an instinct. Where an institution deploys an agent, the institution remains accountable, and the FSB would have a named person within it own the outcome. Singapore says so in terms: engaging a third-party provider does not reduce a firm’s accountability. We take both at face value. It is a sound place to put responsibility, and the sector is better for having it put there.
Notice, though, whom that accountability faces. It is owed to a supervisor, to a board, to the institution’s own governance. It is built, rightly, to hold the institution to account. It does not make the institution anything other than what it is: a counterparty. Not an opponent; we have never thought of the bank or the platform that way. A party, with its own obligations, its own interests, and agents that serve both.
That is the position we hold, and it does not change with the week. The Bespoke AI Confidante stands where the institution structurally cannot: entirely on the client’s side. No institution behind it, no interest of its own, no one to answer to but the person it is formed for. That is what allows a client to meet everyone else in the chain from a position of understanding rather than deference, however well governed the other side has become.
Founder & CEO of SMA Crown Confidential
Bespoke AI Intelligence for Private Decision-Makers
Comments