By Marc A. Rémond
Years ago, visiting Samsung’s headquarters in Korea, I noticed something I did not fully weigh at the time.
Supplier and customer meetings did not happen in the rooms employees used. External visitors were received in spaces set apart from the main office floor, deliberately, as a matter of how the building was organised. A decade in workplace technology later, I understand it was neither. It was governance, expressed in architecture.
I have spent that decade watching meeting rooms grow intelligent. They now sense occupancy, track who is present, transcribe what is said, attribute statements to speakers, and increasingly interpret attention and engagement. The platform analyses what you post under terms you accepted. The voice assistant on the phone in your pocket listens for a wake word you switched on, and earlier this year Apple agreed to a 95 million dollar settlement over allegations that Siri captured conversations through accidental activations, recordings a contractor was hired to review. And the room itself, the one you walked into as a guest, adds its own layer of capture on top.
None of this is covert. That is the first thing the industry tends to get wrong. The fear that animates most writing about ethics in intelligent environments imagines a hidden microphone, a secret listener, a breach to be discovered. It scans the room for a villain. But there is no villain, and that is precisely the problem. You consented to almost all of it. You accepted the building’s terms at the front desk. You accepted the device’s terms when you set it up. You accepted the platform’s terms years ago without reading them. The surveillance, if that is even the right word, is disclosed. It is contractual. And consent, once given, feels like the end of the matter.
It is not, and the reason runs deeper than most of the ethics conversation has reached. Something has changed about the room itself, and our instincts have not caught up. For most of its history the meeting room was a tool. It held a table, a screen, a microphone, and it did what tools do: it sat there until a person used it. We label them, we set policies for their use, we assign them to a department. The intelligent room is no longer that. When a system moves from recording to interpreting, from capturing to deciding who is amplified, who is attributed, what is summarised and surfaced to whom, it stops being furniture and becomes a participant in the meeting. It makes calls. And our governance instincts, every one of them built for tools, do not transfer to a participant. The visitor badge governs a tool. It does not govern an actor that makes decisions about the people in the room.
That is the shift the fear narrative misses entirely. It is still looking for a hidden device when the real change is that the disclosed device started making judgements. And here is the distinction the industry has skipped over as a result: consent is a disclosure event, a single moment, but the room now decides continuously, in every meeting, indefinitely. The room interprets repeatedly. Governance is the only thing that operates on the same timescale as the capability. Consent is a snapshot. Governance is the standing function that matches the room’s standing behaviour. Treating the first as if it satisfies the second is the central failure.
We already know how to do disclosed capture with named accountability, because we do it everywhere else without anxiety. When you join an organisation, you agree to terms governing the monitoring of company systems, and there is a named data controller, a policy, a grievance process. When you buy a concert ticket, the fine print tells you your image may be recorded and used, and there is a venue operator whose name is on that policy. When you register as a visitor and hand over your identification, sometimes signing a form, the organisation receiving you is accountable for what it does with that data. In each case the consent is disclosed and the owner is identified. Someone answers for the consequence.
The AI-enabled meeting room copied the consent and dropped the owner. The visitor consented to being a visitor. They never consented to being interpreted, because no one in the chain decided that they were. The original equipment manufacturer built the capability. The integrator tuned its behaviour for the site. The enterprise deployed it into the room. Three capable parties, and not one of them holds the question of who is accountable when the room acts on what it heard. It said nothing about the intelligence in the room.
What Samsung did, in the practice I observed, was refuse that gap before it opened. By receiving external parties in spaces set apart from the employee environment, the organisation made an explicit decision about where interpretation could occur and who could be subject to it. Someone owned that decision. That is what accountability looks like when it is designed in rather than discovered after an incident.
The leadership implication is not that every organisation must build a dedicated visitor floor. Most cannot, and they should not conclude they are therefore exempt. Spatial segregation is one implementation of the principle, not the principle itself. Samsung had the real estate to solve it with architecture. The same governance is available to an organisation with eight meeting rooms and no spare floor, through mechanisms that cost far less and that already sit inside functions someone owns today.
There are two of them, and they map cleanly onto two different buyers. The first is lobby registration, and it belongs to the facilities leader. Visitor flow, physical access, the front desk and the badge are already theirs. Extending what the badge discloses, so that a guest entering an AI-enabled space is told what the room captures and acknowledges the policy that governs it, is a facilities decision made by the person who already owns the door.
The second is calendar-invite acceptance, and it belongs to the IT leader. In Microsoft 365 and Google Workspace, the invitation already names the location, and a booked room is a resource with its own identity, distinct from an online meeting or an unequipped space. The system therefore already knows whether a meeting lands in an AI-enabled room. An organisation can configure the acceptance of such an invitation to carry a disclosure: this meeting takes place in a room equipped with AI capture, and by accepting you acknowledge the data policy that governs it. The disclosure is delivered at the precise moment of consent, tied to the act of acceptance, owned by the function that already owns the tenant and the room resources. It catches the participant who joins from their own office and would never have passed a front desk.
These are not competing answers to one question. They are addressed to two surfaces and two owners. The facilities leader controls the physical threshold; the IT leader controls the calendar. Sold together, they close the whole gap, the in-person visitor and the remote participant, the lobby and the invitation, with neither leaving the other’s blind spot uncovered. What both share is the thing the room was missing: a named owner who answers for the disclosure. A capability distributed across a chain is a capability no one owns. Accountability does not survive being shared between three parties. It has to land on a role.
For the integrator, this is the part worth sitting with, because it is not a burden. It is a deliverable, and it adapts to whoever is across the table. You cannot sell most clients a separate building for guests. You can build room-tier classification, entry disclosure, and capture-and-retention specification into the deployment as a standard part of the work, and you can pitch it to the owner who controls the relevant surface. Selling to facilities, the answer is lobby disclosure, owned where access is already owned. Selling to IT, the answer is calendar disclosure, owned where the tenant is already owned. Often the answer is both. Governance becomes a line item rather than a building project, and an accountable specification rather than an afterthought the enterprise discovers when something goes wrong.
The strategic insight is this. Ethics in the intelligent room is neither a moral panic nor a luxury reserved for organisations with campuses to spare. It is a governance practice with an implementation tier for every budget and a named owner for every surface. The room that interprets, predicts, and arbitrates is not a tool that captured too much. It is a participant we never decided to govern. The threat is not the listening. It is deploying a decision-maker into the room while mistaking the consent you collected at the door for the accountability you never assigned. The organisations genuinely exposed are the ones that confused having no dedicated space with having no obligation. Name the owner, and the fear resolves into something you can actually build.
