← Back to Articles
The short answer

Classification turns on intended purpose, not on the technology. If a tool is intended for diagnosis, prevention, monitoring, treatment or alleviation of disease, it will generally meet the definition of a medical device and fall under the Health Products Act 2007 in Singapore, with HSA registration requirements following. Administrative, scheduling and documentation tools generally do not. AIHGle 2.0 added worked examples distinguishing the two, because the boundary has caused persistent confusion since 2021.

Intended purpose, not technology

The most common misconception is that classification follows from what the system is built on. It does not. Two tools using the same underlying model can land on opposite sides of the line depending entirely on what they claim to do.

A large language model summarising a consultation for the medical record is doing documentation. The same model flagging that a described symptom pattern warrants urgent referral is doing something closer to triage. The technology is identical. The intended purpose is not.

Where the line usually falls

  • Generally in scope. Diagnosis, screening, risk prediction that drives clinical action, treatment recommendation, image interpretation, patient monitoring with clinical alerting.
  • Generally out of scope. Scheduling, billing and coding, roster management, transcription for the record without clinical interpretation, administrative correspondence.
  • Genuinely contested. Clinical decision support that surfaces guidance a clinician could have looked up, patient-facing symptom checkers, and wellness tools that stop just short of clinical claims.

AIHGle 2.0 added worked examples specifically because that third category generated most of the confusion after the 2021 guidelines.

Why marketing language creates regulatory exposure

Intended purpose is established partly through how a product is described. A vendor whose website promises the tool will detect deterioration earlier than clinicians has made a claim, whatever the contract's disclaimer section says. Institutions have been surprised to find that their own internal communications about a tool contributed to how its purpose would be read.

Our view

The failure pattern is consistent: classification gets treated as a compliance formality to be resolved after the purchasing decision, by whoever owns regulatory. By then the budget is committed and the answer is unwelcome, which creates pressure to reach the convenient conclusion.

Ask it during evaluation instead, and make the vendor state their position in writing. "Do you consider this a medical device in Singapore, and on what basis" is a fair question with a short answer. A vendor who has sold into this market has taken a view. A vendor who becomes vague is either inexperienced or hoping you will not ask again.

The related question worth asking: if you configure or extend the tool, does that change its intended purpose? Institutions that build workflows on top of a purchased system can move it across the line without realising they have become a manufacturer.

Sources

  1. Ministry of Health Singapore, Artificial Intelligence in Healthcare Guidelines (AIHGle 2.0), including worked examples on medical device scope. go.gov.sg/aihgle
  2. Health Sciences Authority, Regulatory Guidelines for Software Medical Devices (GL-04). hsa.gov.sg

Getting the Question Asked Earlier

We build the fluency to raise classification at evaluation, not after signature.

Start a Conversation →