An exemption from manufacturer licensing and product registration, available to public healthcare entities, for Class A and B AI-SaMD in non-critical conditions.
On 13 February 2026 HSA published findings from its public consultation on an AI-SaMD sandbox. It would exempt certain AI software as a medical device from manufacturer licensing and product registration requirements, but only for selected public healthcare entities developing solutions for use in public healthcare. It covers Class A and Class B AI-SaMD that only aim to diagnose or drive clinical patient management for non-critical medical conditions, and comes with conditions including consultant-level clinical oversight, senior leadership endorsement, annual self-attestation to ISO 13485, prior notification to HSA, and telling patients when it is used in their care.
HSA published the findings of its public consultation on 13 February 2026. The proposal exempts qualifying AI-SaMD from manufacturer's licensing and product registration requirements, which are ordinarily the gating steps before a device can be supplied in Singapore.
The rationale is familiar. Public healthcare institutions build tools for their own use, on their own populations, at a scale that does not justify a full commercial regulatory pathway. Requiring one either stops the work or pushes it into an unregulated grey zone.
| Dimension | Scope |
|---|---|
| Who | Selected public healthcare entities, developing for use in public healthcare |
| Risk class | Class A and Class B AI-SaMD |
| Clinical scope | Diagnosis or driving clinical patient management for non-critical conditions only |
| Future scope | HSA indicated it will consider extending beyond public healthcare |
This is not deregulation, and reading it that way would be a mistake. HSA has traded pre-market gatekeeping for named accountability. The registration step goes away and is replaced by a consultant who owns validation and a CEO who signs off. That is arguably a higher bar for the individuals involved, because the responsibility is personal and specific rather than institutional and diffuse.
Which is the interesting part for anyone outside the public sector too. It signals the direction of travel: regulators increasingly want to know who decided, not just what was submitted.
Our practical advice for private providers, who do not qualify today: adopt the conditions anyway. Consultant-level validation oversight, executive endorsement on the record, and telling patients when AI is involved are all defensible practices regardless of whether a sandbox obliges you. If HSA does extend the scope, you will already be operating to it.