Taiwan's Ministry of Health and Welfare opened public consultation this month on draft enforcement rules for the National Health Insurance Data Management Act, the statute promulgated on 19 December 2025 that finally governs the secondary use of one of the most valuable health datasets on earth. The draft — open for comment until 22 June 2026 — establishes an advisory oversight council, standardised data-handling and transmission protocols, and a statutory opt-out for data subjects. It is the closing chapter of a constitutional saga, and on the evidence so far, Taipei has written a proportionate ending.
What the dataset is, and why this matters
Taiwan's single-payer National Health Insurance scheme, launched in 1995, covers more than 99% of the island's 23 million residents. The claims data it generates — the National Health Insurance Research Database — now holds over 70 billion medical records and 3.4 billion medical images, and has underpinned at least 8,414 peer-reviewed publications, according to a 2024 data-resource review in a National Library of Medicine–indexed journal. There is no comparable nationwide, longitudinal, near-total-population health record anywhere in the democratic world. For epidemiology, pharmacovigilance, and increasingly medical AI, it is a strategic asset of the kind most countries can only envy.
That asset, however, was built on shaky legal ground. For years the NHI Administration released de-identified data to researchers under Articles 79 and 80 of the National Health Insurance Act — provisions that never asked the people whose records they were.
The constitutional trigger
On 12 August 2022, Taiwan's Constitutional Court handed down Judgment 111-Hsien-Pan-13, holding those provisions partially unconstitutional. The Court's reasoning is worth taking seriously, because it is the strongest case for regulation here. It found two specific defects: the statute lacked clear rules on the subject, aims, and scope of external data transmission, and it lacked any independent supervisory mechanism; and it gave citizens no way to opt out of having their health records used beyond the purpose for which they were collected. The Court did not order the database shut down. It gave the legislature three years to build the missing guardrails — a deadline that expired in August 2025.
This is the right way to read the ruling: not as a privacy-versus-research zero-sum verdict, but as a demand for informational self-determination and institutional accountability. A dataset this powerful, assembled without consent, needs a control mechanism. The Court said so, and it was correct.
What the draft actually does
The enforcement rules now in consultation operationalise the December statute, and the architecture is notably restrained. Three features stand out.
- An advisory oversight council, not a veto-wielding bureaucracy, to review and supervise secondary-use practices — supplying the independent supervision the Court found missing.
- Standardised handling and transmission protocols, with NHI data de-identified before release and processed only inside designated secure environments. Outputs may not contain anything capable of re-identifying an individual, and commercial, profit-oriented use is expressly prohibited.
- A statutory opt-out. Following the law's commencement, data subjects get an initial 30-day window to withdraw their records from secondary use, and — crucially — they retain the right to opt out at any time thereafter. Unauthorised access to the system carries fines up to NT$10 million (about US$318,000).
This is a default-in, opt-out model, not a consent-each-time model. That distinction is everything. A strict opt-in regime, of the sort some privacy advocates prefer, would have shattered the database's principal virtue — its completeness. Population-level inference depends on the population actually being in the data; selection bias introduced by mass opt-in attrition would quietly degrade every study built on the resource. Taiwan has preserved the asset while giving individuals a genuine, durable exit.
The proportionality scorecard
Where does the draft land on a pro-innovation, evidence-based view? Largely well. It satisfies the constitutional mandate without resorting to the blunt instruments — outright prohibition, hard data-localisation walls, or per-use individual consent — that would have crippled research output. De-identification plus secure-enclave processing is the global good-practice standard, mirroring the trusted-research-environment model used by the UK and the EU's emerging European Health Data Space. Banning commercial exploitation is defensible given the data was compulsorily collected, even if it leaves value on the table for legitimate industry R&D that could be governed rather than forbidden.
Two cautions belong on the record. First, an advisory council is only as good as its independence and its willingness to publish reasons; if it becomes a rubber stamp, the Court's supervision requirement is met in form, not substance. Second, the opt-out's design will determine its credibility — it must be frictionless, well-publicised, and honoured retroactively, not buried in a government portal few citizens visit. The 30-day clock should be a floor on awareness, never a trap.
The wider signal
For the Asia-Pacific, Taiwan is modelling something useful. Faced with a binding privacy ruling over a flagship public dataset, it chose calibrated governance over either defiance or destruction. As Japan, South Korea, and others weigh how to unlock health data for AI without forfeiting public trust, the Taiwanese template — court-mandated guardrails, secure enclaves, real opt-out, dataset intact — is a more replicable answer than the localisation-and-lockdown reflex spreading elsewhere in the region. The consultation closes on 22 June. Taipei should hold its nerve.