Each bank seals its own figure
It compares its figure against a price table the two banks agreed in advance — on its own machine. Only a salted hash of the result leaves the building; the raw number never does.
Inside your bankFor a board, a trade-finance head, or a regulator
Trade-based money laundering hides between institutions: the exporter’s bank sees the invoice, the importer’s bank sees the payment, customs sees the declaration, and no one party sees all three. Comparing them today means pooling a customer’s data across banks — which the law forbids and no bank wants to hold. Mirror lets two banks confirm the figures agree without either revealing the price, the customer, or the amount, and hands each a signed record it can defend to an examiner.
What it does, in three steps
It compares its figure against a price table the two banks agreed in advance — on its own machine. Only a salted hash of the result leaves the building; the raw number never does.
Inside your bankIt compares only the hashes and returns one word: reconciled, divergent, or undetermined. It is built so it cannot read a price, a customer, or a band — there is nothing in its inbox to leak, subpoena, or breach.
The only step outsideEvery answer — including a clear — is cryptographically signed and can be checked by anyone, offline, with the network unplugged. “Prove why you cleared this trade” becomes answerable.
Checkable by anyoneWhat it deliberately will not do
These are designed inabilities, not roadmap gaps — each one is a liability the product is built to be structurally incapable of holding.
It reports facts with their denominator, never a risk score or a probability about a person. A divergence is a reason to look, never an allegation.
It is structurally incapable of holding a suspicious-activity filing, at any depth — so it can never tip anyone off. Your MLRO decides; Mirror hands evidence, not an instruction.
No custody, no settlement, no freeze, no release. It is not in the payment path and holds no account, card, or routing detail anywhere.
The operator holds only hashes. A clear is a defensible reconciliation, not a verdict of innocence — and the honest limits are published, in writing, on every record.
Why this shape, and why now
The regulators are asking for exactly this: a way to compare sensitive data across institutions without pooling it. Regulators already let banks share financial-crime signals; what none of them supplies is a way to do it that privacy law can live with, so the permission goes unused. And the central-utility approach — a pooled database everyone must trust — has repeatedly stalled on precisely the privacy and liability problem Mirror removes by construction. FATF, meanwhile, is pushing privacy-enhancing technologies.
Every claim here is public fact. Mirror claims no regulator’s endorsement, no pilot it has not run, and no registered entity of its own; the fit is a property of the architecture, not a badge.
A legal safe harbour to share, in force since 2001 — and chronically under-used, because it grants permission and specifies no method a bank’s privacy counsel will sign off.
A fresh legal basis for obliged entities to share, PET-agnostic by design. The AMLA/EDPB guidance on how is still unwritten — the exact gap a blind mechanism fills.
A central platform that works because Singapore legislated a safe harbour for an operator that sees everything — a model that meets a hard ceiling the moment it tries to cross a border.
How it is priced
By the corridor — a ratified price table between a pair of banks — per year, split between the two members. Never a fee on the size or number of trades: metering the flow would pay a bank to submit less, and coverage is the whole value. A corridor is the entire rail for that lane, not a per-check charge. The rate card is the parties’ own; Mirror invents no prices.
The honest status
Because it belongs here as much as anywhere: Mirror is pre-revenue and has no registered entity yet. There are no named bank partners, no live production corridors, and nothing on this site is an offer of service. What exists is a complete, tested protocol, an offline-verifiable operator, an on-prem banding tool, and a published threat model — the parts that are hard to fake. The parts that remain are institutional: incorporation, a first design partner, and a regulator-sandbox pilot. A serious counterparty should weigh the engineering against that reality, not around it.
What a pilot looks like
Two banks (or two desks inside one bank, on day one) agree a corridor table, run the on-prem bander against a historical extract with no network, and receive signed verdicts they verify offline. No customer data is pooled, and nothing about the pilot obliges a filing. The output is a small, regulator-ready evidence pack.