For couples after UAE premarital screening: see only the genes you both carry, in plain Arabic or English, with a counsellor one tap away.
The two phones compare results privately. Neither partner learns any other gene the other carries, and nothing reaches either family unless you choose to share it. If you wish, relatives can be told that a test may help them, through your counsellor if you’d rather not be named.
1 in 13
couples in Dubai’s first 1,000 screened carried the same gene (79, or 7.9%, including alpha-thalassaemia)
Nature Medicine 2026565
of 11,748 Abu Dhabi couples screened 2022 to July 2026 had an increased chance (4.8%)
DoH, via Gulf NewsToday one partner’s carrier status can travel to the other, and to both families, before anyone has explained that being a carrier is not a disease. The challenge asks for the opposite: no exposure of one family member’s information without permission.
Abu Dhabi has trained hundreds of physicians to counsel because specialists are scarce. Every couple still needs the same careful explanation, in their language, at their reading level.
In consanguineous families relatives share variants, yet in one US study only 6% of first-degree relatives went on to be tested. People hesitate to name themselves.
For this pairing only, the lab turns each gene a person carries into a token only it can make. Nobody can invent a gene they don’t carry or reuse someone else’s.
Each phone scrambles its tokens with a secret it never shares, pads the list so its length reveals nothing, and sends it through the relay, encrypted.
Each phone adds its own lock to the other’s list, with a proof that it locked honestly. Items both carry end up identical; everything else stays random noise. The lab then confirms the match from its own records and names only the shared genes.
This is authorised private set intersection (De Cristofaro and Tsudik), with the lab as the trusted issuer: a partner can’t probe by pretending to carry extra genes, the problem Baldi et al. (2011) raised. Abu Dhabi’s Responsible AI Standard names secure multi-party computation as a privacy technique; this is one.
Each partner
Learns: The genes they both carry, explained
Never learns: Any other gene the other carries, or how many
The relay server
Learns: That two phones exchanged three encrypted messages
Never learns: Any gene, any token, or the result
The lab
Learns: That this couple paired, and the match it confirms (it already holds both results, as today)
Never learns: Anything new about either person
A counsellor
Learns: Only the items a person ticked when booking
Never learns: Anything after consent is withdrawn
Relatives
Learns: A gene and their own chance of carrying it
Never learns: Who sent it, when a counsellor passes it on
The health system
Learns: Counts: understanding, uptake, time to referral
Never learns: Any individual, or any marriage decision
Chances are drawn the way a counsellor sketches them: four pregnancies, one affected, two carriers, one neither. Every explanation comes in simple, standard and detailed versions, in Arabic and English, and says plainly what is risk and what is diagnosis. The words avoid “incompatible”; the couple is “a couple with an increased chance”.
SharedCongenital adrenal hyperplasia · CYP21A2
3 in 4 children will not have CAH.
In each pregnancy: 1 in 4 that the child has the condition, 2 in 4 a healthy carrier, 1 in 4 neither.
Treatable with daily medicine and found by newborn screening. If you know of the chance, tell the birth team: a salt-wasting crisis can come before the screening result is back.
Go ahead, test in pregnancy, IVF with embryo testing, reconsider plans together, kafala or no children, or talk first. Equal weight, no advice, and one plain line on costs and cover under them all.
An AI explainer answers only from vetted facts, has its numbers checked against them, refuses to probe a partner or decide for anyone, and hands distress to a person the same day. Anyone can ask a person instead.
Booking shares only what you tick. The desk tracks time to first contact, referral and documented closure, with a tamper-evident audit trail.
Sealed notes tell a cousin or sibling their own chance of carrying the same change. The service can’t read them or tell who sent them, and a counsellor can pass one on so no name is attached.
We never count marriages called off. The challenge says success must not rest on any single personal choice, and a couple’s decision is theirs.
The DoH premarital panel results already flowing through Malaffi; UAE Pass sign-in; the 22 screening centres and their trained physicians.
Demo: Vercel and Amazon Bedrock in Sydney, synthetic data only. Production: hosted in the UAE, as Decree-Law 49/2023 and the health data law require. Amazon Nova Pro is offered in AWS’s UAE region (not yet tested there), and the safety layer works with any model.
First the DoH sandbox, ethics approval, a DPIA and an independent crypto audit. Then a randomised pilot of the explainer and family notes with couples newly found at increased chance, against usual care. Months 18–30: lab tokens and Malaffi at three or four centres. A licensed provider operates it; DoH sees only counts.
We are looking for a genetic counsellor in the UAE to work with us, starting with the phase-0 sign-off of every condition card, fact and Arabic string before any real use.
First cousins who share a gene, an unrelated couple who share none, double first cousins who share one full and one milder match, and second cousins whose two milder changes are shown as a low-concern match. All synthetic, all running the real protocol.
Try it with two phones