HSA SaMD & AI Device Landscape — What's In Force in Singapore Right Now (April 2026)

In the space of four months — December 2025 through March 2026 — HSA Singapore has reshaped how Software as a Medical Device (SaMD) and AI-enabled medical devices are regulated. Four distinct instruments are now in force or recently effective, and a fifth (the AI-SaMD Exemption Order) has created an explicit two-tier regime that lower-risk public-sector developers benefit from while private-sector developers do not.

If you're building, registering, or maintaining SaMD in Singapore, this is the current-state map.

TL;DR — what's live

These are not proposals. They are current obligations.

The three-document web for SaMD lifecycle

For any SaMD registered in Singapore, the operational burden now comes from three documents that must be read together:

GL-04 R4 — the lifecycle spine

GL-04 R4 is the primary lifecycle guidance. It covers the full software medical device lifecycle — development, verification, deployment, post-market surveillance, cybersecurity, continuous improvement — aligned with IMDRF and ICH principles. Effective December 31, 2025 means any dossier submitted or updated after that date must reflect R4 expectations. Dossiers that still cite the earlier revision are, in practice, non-compliant.

GN-37 — the Change Management Program (CMP) pilot

GN-37 introduces a formal CMP pathway for SaMD — including ML-enabled SaMD — letting manufacturers push pre-authorised model updates without triggering a full re-registration each time. The pilot is labelled "optional," but the operational reality is different:

• Teams that don't engage the CMP face re-registration delays every time they update an ML model.
• Teams that do engage the CMP commit to documented change control, risk assessment per change, and audit trails.

For any SaMD company actively iterating its ML models, the CMP is effectively mandatory if you want to ship updates at a reasonable cadence.

GN-21 — the change-notification framework

GN-21 governs the formal change-notification process for registered devices. Read alongside GL-04 and GN-37, it closes the loop on what goes into a notification vs what stays internal to CMP vs what triggers full re-submission.

The practical rule: any SaMD company without a dedicated RA resource is at risk of compliance drift against this three-document web. The documents aren't contradictory, but they're not trivially mergeable either.

AIHGle 2.0 — the clinical-AI grey zone is closed

The AI in Healthcare Guidelines version 2.0, co-published by MOH and HSA on March 10, 2026, is the forcing function for clinical AI tools that have been operating in a grey zone.

Previously, a hospital innovation team deploying a clinical AI tool could reasonably defer the question of whether the tool was regulated as a medical device. AIHGle 2.0 closes that gap. The guidelines:

• Provide explicit worked examples of when clinical AI crosses the line into AI-SaMD regulatory scope.
• Draw attention to AI-SaMD regulatory triggers — intended use + therapeutic purpose + output relied upon for clinical decisions.
• Align MOH's clinical governance expectations with HSA's device classification.

The operational impact: any clinical AI tool used in a Singapore healthcare setting must now explicitly evaluate whether it triggers AI-SaMD classification. The "we're not a medical device" assumption no longer survives AIHGle 2.0's examples.

GN-15 R13 — CDSS is in scope

GN-15 R13 is the current (March 2026) revision of HSA's Medical Device Product Registration guidance. The material update for 2026: Clinical Decision Support Software (CDSS) is now explicitly noted as in scope for device registration.

This is a meaningful change. A class of products that previously operated under "it's just decision support, not a device" positioning now need to do a formal classification review:

• If your CDSS influences a specific patient's treatment decision, it almost certainly triggers device regulation.
• If your CDSS surfaces general clinical knowledge (lookup, reference, dosing calculator without patient-specific inputs), you may still fall outside device scope — but the burden is now on you to document that reasoning.

Any CDSS product targeting Singapore healthcare that hasn't been formally classified should do that review immediately.

The AI-SaMD Exemption Order — a two-tier regime

The Health Products (AI Standalone Mobile Application — Exemption) Order 2026 took effect February 13, 2026. It exempts lower-risk AI-SaMD (typically Class A/B) developed by selected public healthcare entities under the oversight of a senior clinician.

Read the Order carefully: it is explicitly scoped to public healthcare. Private-sector AI-SaMD developers — healthtech startups, private hospitals' innovation teams, commercial AI vendors — are not exempted. They must navigate full registration via the standard pathway.

The commercial implication matters: a private-sector AI-SaMD vendor trying to win contracts in public hospitals is now competing against public-institution-developed AI-SaMD that enjoys a lighter regulatory burden for the same risk class. This is a deliberate policy choice — it accelerates innovation within the public healthcare system while keeping the safety rail up for commercial entrants — but it is also a real commercial asymmetry worth naming in a compliance discussion.

Baker McKenzie's February 2026 analysis and Allen & Gledhill's summary both confirm this scoping. Pacific Bridge Medical notes the exemption applies specifically to Class A/B AI-SaMD overseen by a senior clinician at a public institution.

What SG medtech teams should do now

If you have a registered SaMD in Singapore

1. Audit your current dossier against GL-04 R4. If cited revisions or references are still pointing at the earlier GL-04, that's an immediate remediation item — not a next-submission item.
2. Decide on CMP engagement. If you update ML models or software features at any meaningful cadence, you need the CMP pathway. Start the engagement process now rather than waiting until your first post-market change.
3. Re-read GN-21 alongside GL-04 and GN-37. The three-document web is the current operating baseline.
4. Cross-check your post-market surveillance (PMS) plan — GL-04 R4 assumes a live PMS function, not a binder on a shelf.

If you develop clinical AI tools or CDSS

5. Run the AIHGle 2.0 worked examples against your product. If your tool crosses the line into AI-SaMD per the guidelines' examples, start the classification process.
6. Review CDSS scope against GN-15 R13. Document your classification reasoning — particularly if you're claiming out-of-scope status.

If you develop AI-SaMD in the private sector

7. Do not assume the AI-SaMD Exemption Order applies to you. Unless you are a public healthcare entity, it does not.
8. Plan full-pathway registration under the current HSA framework. The AI-SaMD Exemption Order is not a deferral mechanism for private sector — it is an explicit carve-out for public sector only.

If you deploy AI tools in hospitals (non-medtech)

9. Check AIHGle 2.0's coverage of your deployment context. Hospital innovation teams are an increasingly common audience for HSA guidance, and AIHGle 2.0 is explicitly aimed at you as well as vendors.

Why the alignment matters

The four HSA documents (GL-04 R4, GN-37, AIHGle 2.0, GN-15 R13) are not independent — they are coordinated outputs of a single regulatory alignment effort. Singapore is deliberately moving its SaMD and AI-device regulatory posture into closer alignment with IMDRF, FDA's AI-SaMD lifecycle direction, and the EU's MDR / AI Act combined framework.

Operationally, that means:

• Compliance work has compounding value. A company that builds its SaMD lifecycle documentation correctly for HSA will largely have the base artefacts required for FDA TPLC, EU MDR, and Australia's TGA Compliance Principles (which named SaMD as an enforcement priority for 2026–2027).
• Falling behind compounds the other way. A company that didn't update against GL-04 R4 in January has accumulated four months of drift. Multiply across jurisdictions and the remediation cost scales.

Sources

• GL-04 R4 (HSA): Regulatory Guidelines for Software Medical Devices — Life Cycle Approach (Dec 2025)
• GN-37 (CMP for SaMD including ML-enabled SaMD): HSA guidance-documents index
• AIHGle 2.0 (MOH + HSA, Mar 10 2026): Baker McKenzie summary (primary HSA press release also available; check HSA's announcements page)
• GN-15 R13 (Medical Device Product Registration, Mar 2026): HSA guidance-documents index
• AI-SaMD Exemption Order 2026 (eff. Feb 13 2026): Allen & Gledhill Health Products (AI Standalone Mobile Application — Exemption) Order analysis; Baker McKenzie Feb 2026 note on public-sector scoping; Pacific Bridge Medical analysis

Related regMD coverage

• HSA Singapore Publishes GN-02 Revision 8 — Updated Licensing Rules for Manufacturers, Importers & Wholesalers (April 2026)
• TGA Compliance Principles 2026–2027 — SaMD Is Now an Explicit Enforcement Priority
• Singapore HSA AI-SaMD Sandbox Readiness
• Private Medtech AI-SaMD — Singapore HSA Pathway
• Medical Device SaMD — Software Classification

This article summarises publicly available HSA, MOH, and law-firm guidance as of April 17, 2026, and does not substitute for regulatory or legal advice. Always refer to primary HSA source documents and consult your Regulatory Affairs adviser for device-specific decisions. The regulatory landscape for SaMD and AI-enabled medical devices is evolving rapidly; check current revisions before acting on this summary.