Ong Ye Kung: Requiring Consent for Every Data Point Would Make Healthcare AI a 'Non-Starter'

Singapore's most ambitious healthcare AI project is also its most sensitive, and Parliament heard on Aug 5 exactly how the government plans to navigate the consent question. Health Minister Ong Ye Kung told the House that requiring explicit patient consent for every piece of data — even anonymised data used to train medical AI — would make the entire effort "a non-starter," drawing a clear line on how the Singapore Medical Foundation AI Model (SIMFONI) will get the local data it needs.

Replying to Non-Constituency MP Andre Low, Ong said SIMFONI — announced on Jul 9 — will train AI tools on de-identified, anonymised patient data to help clinicians diagnose chronic conditions such as diabetes, high cholesterol and eye disease. The minister said the de-identification process began years ago and that there have been no data leaks so far. MOH will start with a couple of specialty areas in the public healthcare sector, and if successful, expand to more specialties and possibly private practitioners.

SIMFONI exists because most healthcare AI models are trained on Western populations, which limits their accuracy in Singapore's clinical settings — a gap with real consequences, since Asians develop conditions like diabetes at lower body mass indexes than Westerners. The programme sits under the Consortium for Clinical Research and Innovation Singapore (Cris), bringing MOH research programmes together around a common goal: models built on local data, validated against Singapore's own clinical guidelines.

Ong also pushed back on the hype surrounding the technology. Responding to Yip Hon Weng's question on automation bias — the risk that doctors gradually defer to AI recommendations — he warned against treating AI as "a hammer going around looking for nails," and stressed that deployment must follow clinical protocols with clinicians making the final judgment. "All this hype will subside to a more realistic level," he said, adding that MOH would rather start sober and judicious than wait for that correction.

Why it matters for Singapore: The exchange settles a fundamental policy question: Singapore will not let per-patient consent requirements starve its medical AI of training data, but will lean on anonymisation and clinical oversight instead. For healthcare providers and AI developers, that signals a permissive-but-guarded data environment — and for patients, it means the safeguards that matter will be technical (de-identification, audit trails) rather than procedural (consent forms). The next test is whether MOH's promised safeguards keep pace with the models.