In June, the Financial Stability Board set out twelve sound practices for its adoption and conceded that reviewing an autonomous system's decisions one by one does not scale. In September, the UAE Cabinet approved a Cabinet AI Advisor built on thirty-two specialised assistants. The question for every board in this region is no longer whether the machine is in the room, but where human judgement sits once it is.
The conventional answer is “human in the loop”. It asks someone to approve what the system has already concluded, without seeing how it reached that conclusion and, in many organisations, without the authority to reject it. This does not restore judgement. It relocates blame to whoever touched the system last, while the board’s accountability stays where it was.
This region has an older and, in my opinion, better model for how counsel and decision should relate. The majlis, inscribed in 2015 by the UAE, Saudi Arabia, Oman and Qatar on UNESCO's list of intangible cultural heritage, is the space where a community gathers to resolve its problems. It is open to all. Judges and sheikhs adjudicate disputes in the open. The host presides, listens, decides, and bears the cost of the gathering. The late Sheikh Zayed bin Sultan Al Nahyan, Founding Father of the UAE, governed through regular majlises where everyone who spoke was heard.
I write about it as a guest of this country, not as a custodian of its traditions. But any risk practitioner recognises its logic: consultation without ownership is noise, and ownership without consultation is blindness.
The lesson for AI governance is what the majlis asks of those who enter. Open to all does not mean open on any term.
Everyone admitted is known and greeted, is bound by the etiquette of the room, speaks in front of all, and answers for what they say. A black-box system fails that test by design: it offers conclusions with no face, no stated reasoning and no reputation to lose. It can be admitted, but only on conditions.
Regulators have reached the same place by another road. The FSB suggests firms may need to treat AI agents as “synthetic employees”, with identities, access rights and assigned responsibility, and for agentic systems it favours “human in command”: oversight of boundaries and guardrails rather than approval of every output. In majlis terms, the conditions of admission are three. The machine enters with a named human who answers for it. Its reasoning is laid out in the open (inputs, retrieved context and logs included), so that people who understand the subject can question it. And the room keeps the standing to refuse it.
This is where trust, fairness and risk meet, and it is why the majlis resonates with me: they are what a career in regulated finance teaches you to protect. Trust in a majlis is built by presence and by reasoning spoken aloud before people who can answer back; it cannot be delegated to a model, only rebuilt around one. Fairness means every voice is heard; a model hears only what is in its data, so the voice it silences is the one nobody notices is missing. Risk is managed in the majlis as it should be in a firm: weak signals surface because people meet and speak, and disputes are settled before they become losses. The host does not check every sentence. He decides who sits, what is discussed, and what is done.
For boards and executives, this is the design task. Call it the strategic majlis: the body that owns the decision. Let the machine ingest, synthesise and recommend at a scale no committee can match; that is what it is for. The decision itself is taken in the room, in the open, by people who answer for it.
A majlis, finally, is not a room. It is the people who sit in it. Review protects anyone only when the reviewer could have done the work themselves; a strategic majlis staffed by people who cannot interrogate the guest is a ceremony, not governance. The investment that matters is in people with the domain depth and the standing to say no, and in leaders willing to listen last.
The machine may speak in the majlis. It may not preside.
(This is a contributed article by Roberto d’Ambrosio CEO, Axiory)
