"An AI manages my flat" sounds, to most London owners, like a risk — and if we are honest, it should. Software that can message your guests, reprice your nights and touch your money is only as trustworthy as the thing that can stop it. So before you meet any of our AI systems that do things, meet the one whose entire job is to prevent them: Themis, our compliance guardian. Themis is the reason we can run a property business on AI without asking you to take anything on faith. Its only power is refusal. It cannot list, price, message or pay — and, uniquely on our team, it cannot be overridden by anything else we have built.
The AI whose only power is to stop things
On our public /team page, every AI system carries a card stating what it does and — just as prominently — what it cannot do. Themis's card reads, verbatim: "Checks every action before it happens; anything irreversible — a payment, a message, a contract — is stopped and handed to a person." And under what it can't do: "Can't be overridden. This is the deny-by-default rule at the centre of everything." That second line is the important one. Most AI products advertise capability; ours leads with a constraint that no other component of the system can route around.
Deny-by-default means the burden of proof sits with the action, not the objection. An action is not allowed because nothing forbade it — it is blocked unless it is explicitly permitted, and anything irreversible never gets automatic permission at all. It stops, and a named human decides. This is the opposite of the usual automation posture, where software acts freely and humans clean up afterwards. We think the order matters more than the technology: for your flat, your guests and your money, the mistake you cannot undo is the only mistake that really counts.
One gate, ten systems
Themis is not a marketing flourish bolted onto one product — it is the centre of a team of ten named AI systems across six departments, each disclosed as an AI system, never a fake person. Seven are live today; three are labelled "in rollout", which means exactly what it says: the capability is coming online and we will not claim it in the present tense until it is. Every card names the system's hard stop. Here is a sample, quoted verbatim from the /team page:
| System | Role | Status | What it cannot do (verbatim) |
|---|---|---|---|
| Themis | Compliance guardian | Live | Can't be overridden. This is the deny-by-default rule at the centre of everything. |
| Solon | Open-book accountant | Live | Can't move money — every payout is approved by a human. |
| Hephaestus | Turnover orchestrator | Live | Can't dispatch a real person or issue a door code alone — a human approves it. |
| Herald | Owner research | Live | Can't contact anyone — outreach is only ever sent by a named human. |
| Kairos | Dynamic pricing | In rollout | Can't push a live price change without a bounded rule and human sign-off. |
| Iris | Guest concierge | In rollout | Can't message a guest on its own — a person sends every reply. |
Notice the pattern: the "cannot" lines all converge on the same gate. Money, messages, contracts, door codes, outreach — every irreversible path runs through a person. That is not ten separate safety features; it is one rule, enforced by Themis, expressed ten ways.
"The most useful thing an AI can do with an irreversible action is refuse it. Speed is cheap; accountability is the product."
The proof the gate is real: the claim we retired
Any agency can publish a governance diagram. The evidence that ours actually bites is public and slightly embarrassing, which is rather the point. Until 20 April 2026 our site carried an occupancy claim asserting an unbroken run of booked nights. When we audited our own evidence, we found we could not produce a continuous calendar export proving zero un-booked nights across the whole period. So the claim was retired on 2026-04-20 and softened to "~87% occupancy since launch" — the tilde is deliberate — sourced to an Airbnb calendar export covering May 2025 to April 2026, supplied on request. We also published the rule for reinstating the stronger claim, so you can see exactly what evidence would be required to bring it back.
That retirement lives on our public claims ledger at /transparency/claims, where every marketing number on this site is listed with its source and a "last verified" stamp. The ledger carries two standing commitments: every entry is re-verified quarterly, and if you find a claim we cannot substantiate, we retract or cite within 48 hours — email compliance@bigbensuite.co.uk. A gate that only ever waves things through is theatre. Ours has a publicly visible casualty.
Where the human sits
The /team page ends with a paragraph we regard as the contract behind everything above it: "These are AI systems, not people — we tell you plainly and would never pretend otherwise. Every payment, message, contract and deployment is approved by Guray Uzun, the founder. The AI is the speed; the human is the accountability." That is a named human at a registered company — Big Ben Suite Ltd, company no. 15855127 — not a support queue. When Themis stops something, it lands on his desk, with his name attached to the outcome. "Run by AI. Answerable by a human." is not a strapline we retro-fitted; it is the operating order of the whole system.
How to audit us in ten minutes
- Read the /team cards and check that every system — including the ones in rollout — names a hard stop, not just a capability.
- Open the claims ledger at /transparency/claims and pick any number from this site; it should appear there with a source and a last-verified date.
- Find the softened occupancy claim on the ledger and read the published reinstatement rule — the record of a claim we withdrew ourselves.
- If anything fails that test, email compliance@bigbensuite.co.uk and start the 48-hour retract-or-cite clock.
