Skip to content

Turn data that can't be shared into decisions no one could make alone.

DoubleBlind matches records across organizations' separate datasets in real time, while every record stays encrypted.

Something that looks new to you is already known somewhere else.

The most valuable data an organization holds is the data it isn't allowed to use. Contracts, regulation, competition and national borders keep each dataset in its own environment, so every decision is made with partial information.

Fraud is a shared problem.

The same face opens accounts at several banks, and each one sees only its own piece.

The evidence can't be pooled.

Faces and ID documents are biometric data, and whoever holds the pool becomes the target.

Workarounds miss what matters.

Flags and file fingerprints catch exact copies, and AI now makes altered photos cheap.

$3.2B
lenders' exposure to synthetic identities in H1 2024. TransUnion, H2 2024 State of Omnichannel Fraud
680,000+
check-fraud reports filed by US banks in 2022, about twice 2021. FinCEN alert FIN-2023-Alert003
71%
rise in one UK insurer's detected fraud from 2024 to 2025, including AI-edited photos. Admiral

How DoubleBlind works

A check goes out encrypted, each organization searches only its own records, and only the permitted answer comes back.

  1. Encrypt

    The query is encrypted with fully homomorphic encryption, so it can be searched while it stays encrypted.

  2. First pass

    CyborgDB's compressed index narrows millions of records to a few, with minimal leakage.

  3. Full match

    An exact encrypted comparison on those few, fuzzy fields included.

FHE is usually too slow for live decisions. This isn't.

<250 ms
p90 across 4 environments of 100M records each
<120 ms
unloaded latency at 100M records per environment, and under 70 ms at 25M
90+ q/s
on a single AWS c8g.24xlarge at 25M records per environment

Measured with one 96-core AWS Graviton4 machine per environment, at recall of at least 0.99.

Models supply intelligence.
DoubleBlind supplies evidence.

Every organization can buy the same detection models. The evidence across everyone the company in the middle serves is the one thing no competitor and no model can reproduce.

Compared while encrypted.

Records are matched while encrypted, in each organization's own environment.

Only the permitted answer returns.

A check discloses only what it's set to: a match, a count, or which fields agree.

Similar, not only identical.

A different photo of the same person, or a misspelled name, still matches.

Real time, thanks to CyborgDB.

Its patented index narrows the search before the detailed comparison.

Runs on standard hardware.

No specialist servers. It deploys as one program and takes your own face or image model.

Feeds how decisions are already made.

Use results as signals in rules, review queues or models, and recheck earlier decisions as new records arrive.

Where other approaches fall short

ApproachWhere it falls short
Pooling the dataBecoming the honeypot that organizations and regulators won't allow
Sharing risk scores, flags and file fingerprintsCatching exact copies and missing a different photo, a misspelled name or an edited image
Ordinary search with encrypted storage or trafficThe search still reads the data in the clear
Secure enclavesTrusting whoever runs the enclave
Federated learningSharing what a model learned, not whether two records match
General-purpose encryption librariesBuilding the indexing, coordination and result controls yourself, at speeds too slow for live decisions
Data-collaboration toolsBatch analytics, not live matching

Built for platforms that serve many organizations.

DoubleBlind is for an organization that serves many others whose data must stay separate, and is trusted to check across them: onboarding, payments, claims, hiring, eligibility.

Identity verification

“The same face opened accounts at three banks under three different names.”

Who runs the checkIdentity-verification platforms; interbank and gaming schemes

Bank fraud and anti-money-laundering

“The mule account opened here was flagged at another bank last week.”

Who runs the checkInterbank utilities; anti-money-laundering platforms

Insurance claims

“The same damaged-car photo was filed with two insurers.”

Who runs the checkClaims platforms; claims-data operators

Healthcare record matching

“The right patient, matched across hospitals that can't merge records.”

Who runs the checkHealth information exchanges; record-linkage vendors

Up to half of patient records fail to match between hospitals. Pew Charitable Trusts, 2018

Hiring fraud

“Today's new hire used the same face at three other companies.”

Who runs the checkHR and background-check platforms

Aid de-duplication

“The same person registered for aid twice, in two agencies' systems.”

Who runs the checkAid networks

Fully homomorphic encryption, on CyborgDB's patented index.

DoubleBlind uses fully homomorphic encryption, a way of computing on data while it stays encrypted. It is built on CyborgDB's patented index, which narrows the search before the detailed comparison. Cyborg holds 14 issued US patents.

Become a design partner.

Tell us what you need to match, and across whom.