The Ministry of Health and the Health Sciences Authority refreshed Singapore's AI in healthcare guidance in March 2026. Here is a plain reading of what changed.
Singapore's MOH and HSA launched AIHGle 2.0 on 10 March 2026, replacing the 2021 framework. It is guidance, not new legislation. The substantive change for most healthcare organisations is that accountability is now split explicitly across three roles, developer, deployer and user, and if you buy and run AI in your institution you are the deployer. That role carries duties you cannot outsource to your vendor.
The Ministry of Health and the Health Sciences Authority co-developed the revised Artificial Intelligence in Healthcare Guidelines and launched them on 10 March 2026. MOH circular MOH-MHC-0024-2026, dated 13 March 2026 and signed by Adjunct Professor (Dr) Raymond Chua, Deputy Director-General of Health, Health Regulation, addressed it to all HCSA licensees and all healthcare professionals.
The guidelines were developed with Synapxe, the Infocomm Media Development Authority, the Personal Data Protection Commission, the Academy of Medicine Singapore, the College of Family Physicians Singapore, and a range of professional boards and associations. That list matters. It signals a cross-agency position rather than one regulator's view.
The circular lists three key updates:
AIHGle 2.0 also addresses developments the 2021 framework did not anticipate, generative AI among them, and states that AI should augment and empower healthcare professionals with patients at the heart of care delivery.
The accountability split is the part most likely to change how a hospital or clinic operates. Three roles, and most institutions occupy more than one.
Builds the AI solution. Typically your vendor, but an institution that builds its own model in-house takes on this role too.
The healthcare organisation that puts the solution into service. This is where most hospitals, clinics and health systems sit, and where the governance duties concentrate.
The healthcare professional using the tool in practice. Their duty depends on being told enough to exercise judgment, which is why the transparency guidance sits alongside this.
If you are the deployer, the guidance assumes you can answer questions about the solutions running in your institution: what they do, on what data they were built, how their performance is monitored after go-live, and who is accountable when output is wrong.
MOH published two supporting pieces alongside the guidelines: a deployers' toolkit aimed at new adopters, covering internal governance and how to select an appropriate AI solution, and an infographic summarising roles and responsibilities across the three stakeholder groups. Both are available with the guidelines at go.gov.sg/aihgle.
AIHGle 2.0 is guidance rather than legislation, and reading it as optional would be a mistake. It complements HSA's Regulatory Guidelines for Software as Medical Devices, and AI that meets the definition of a medical device stays regulated under the Health Products Act 2007. One of the useful additions in 2.0 is worked examples that distinguish AI solutions regulated as medical devices from those that are not, which is the ambiguity that has caused the most confusion since 2021.
MOH also describes AIHGle as a living document that will be updated periodically. Treating compliance as a one-off exercise against the March 2026 text will not hold.
The deployer role is where this guidance will bite, and it is the role most institutions are least prepared for. A vendor can hand you a model card and a validation study. Only you can decide whether that evidence is good enough for your patient population, and whether the monitoring you have in place would actually catch drift before a clinician does.
The gap we see in practice is rarely legal. Institutions can read the guidance. What they struggle with is having enough people in the room who can tell whether a vendor's answer is a real answer. Working out whether post-market monitoring is adequate, or whether a model trained on a different population will hold up on yours, is a judgment call that needs enough fluency to ask the second and third question, not just the first.
Our practical advice: before your next AI procurement, write down which of the three roles your organisation occupies for every AI system already running. Most leadership teams cannot complete that list, and the exercise of trying is more useful than any policy document you could write.