Bran (Brandon) Myers
Public-Interest Investigation · Global Footprint · June 2026

Who Actually Uses Splink — And What They Do With Your Records

Part four. After The Quiet Joining-Up, Statistics or Operations?, How Splink Works and Ruled Unlawful Next Door, one question was left: who else, and what have they done with the data? I profiled fourteen adopters to the same depth as the UK — then had each profile adversarially fact-checked.

Tier 1 — proven operational

First-party government docs · 3
🇬🇧 UK — Ministry of JusticeCourts, prisons, probation + real-time Core Person Record; daily PNC→police feed
🇬🇧 UK — Office for National Statistics2021 Census ↔ tax / benefits / NHS, near-total population
🇦🇺 Australia — Bureau of Statistics2025 Person Linkage Spine: Medicare + Centrelink + Tax

Tier 2 — listed only on Splink’s own “Used By” page

Self-reported, unaudited · 5
🇬🇧 UK wider public sectorDfE, CMA, Welsh Revenue, councils
🇬🇧 NHS EnglandLive spine is deterministic — Splink in development
🇨🇦 Canada — Ontario MCCSS · ECCCSplink unnamed in the paper
🇩🇪 Germany — DestatisOne uncited line
🇪🇺 EU — Medicines AgencyAsserted
🌍 UNHCRPublic code uses fastLink, not Splink

Tier 3 — pilot · evaluation · research

Not production on real citizen data · 6
🇺🇸 US — Indiana DoHPilot, synthetic data
🇺🇸 US — Florida Cancer RegistryEvaluation, simulated data
🇦🇺 AIHWEvaluating; still on SAS “DALI”
🇱🇦 Laos — child health recordResearch study
🇨🇱 Chile — Ministry of HealthResearch: migrant immunisation
🇬🇲 Gambia — 2024 censusAsserted (uncited)

Checked & NOT Splink

Benchmarked & rejected, or never used · 4
🇺🇸 US Census BureauUses BigMatch — evaluated Splink, rejected it
🇺🇸 NIH “All of Us”Uses Datavant tokenisation
🇨🇦 Statistics CanadaUses its own G-Link
🇦🇺 State linkage unitsChoiceMaker · DLS3 · FEBRL
Documented Splink adopters, tiered by what primary sources can prove. A maintainer’s logo-wall is not proof.

↗ Read the full 14-profile master analysis (GitHub)

The honest answer is narrower than the logo-wall — and that is the point

Splink's maintainers, the UK Ministry of Justice, keep a public “Used By” page. It is the reason this map can exist at all — and it is also a trap, because it is unaudited. When you check each name against the organisation’s own documents, the picture shrinks to something far more honest than a vendor’s logo-wall. So I tiered every adopter by what can actually be proven, and corrected my own earlier overreach in the process.

Independently proven, operational on real population data — 3. UK Ministry of Justice (the sharpest case anywhere), UK Office for National Statistics, and Australia’s Bureau of Statistics. First-party government documentation, not a listing.

Listed only on Splink’s own page — 5. The wider UK public sector, NHS England (whose live system is actually deterministic — the Splink work is still in development), Canada (Ontario + Environment Canada), Germany’s Destatis, the European Medicines Agency, and UNHCR. Real linkage in most cases, but the Splink attribution is self-reported — and in checked cases wrong.

Pilot, evaluation or research — 6. US (Indiana pilot, Florida evaluation, both on synthetic data), Australia’s AIHW (still on its own SAS tool), Laos (a study on children’s records), Chile (migrant immunisation), and the 2024 Gambia census (asserted, uncited).

Checked and NOT Splink — 4. The US Census Bureau evaluated Splink and rejected it for its own BigMatch; NIH’s All of Us uses Datavant; Statistics Canada uses G-Link; Australia’s state linkage units use their own engines. Said plainly, so the rest stays honest.

What they actually do with the data

The UK Ministry of Justice is the one place the tool has crossed from statistics into operations: a real-time Core Person Record that assigns a live cross-system identifier as records are created, court use that retrieves a person’s probation history while preparing a case for sentence, and a pilot that sends probation supervisees’ police numbers to the police every day to check for arrests. Everywhere else, the documented use is statistical — the census, research datasets, de-duplication — or a pilot. The structural danger the MoJ proves is that the same library does both: the research build and the operational build are one configuration apart.

The pattern that holds in every country

Whatever the tool, the pattern is identical: linkage proceeds without consent because the law was written to allow it. Public task in the UK, the Census and Statistics Act in Australia, statutory census duty in Germany, health-privacy carve-outs in Canada. The protection is never the individual’s choice — it is statutory confidentiality, de-identification after the join, and the “Five Safes.” The defence “but it’s lawful” is the indictment: you weren’t asked, you weren’t told, and there was no way to opt out — and that was the design.

What I will not claim

This is not a global conspiracy — these are independent bodies sharing one open-source library, and many do not even do that. It is not mostly illegal — almost all of it is lawful by statutory design, which is the point, not a gotcha. And I will not present a self-reported listing as audited fact: Germany, the EMA, the Gambia and UNHCR entries are flagged as exactly what they are. The strongest, undismissable finding is the one that survives every hostile read — population-scale, person-level record linkage, lawful, and disclosed to almost no one.

Method: 14 adopters each profiled to UK depth, then each profile put through an adversarial verification pass against primary sources; the verifier’s corrections applied strictly. Full profiles, sources and the still-uncorroborated list: master analysis on GitHub.

← All Writing