Key points
- Each AI tool in the clinic, from the scribe to the front desk drafts, gets a row on one register with an owner and a human check.
- Singapore's AIHGle 2.0, published by MOH and HSA in March 2026, asks healthcare organisations to keep a registry of every AI solution they deploy, with its risk classification.1
- Human oversight is mandatory for all Clinical and Clinical-Ops use under AIHGle 2.0, matching the WHO principle that humans stay in control of medical decisions.12
- A small clinic can keep the register on one page and review it monthly.
- wai drafts and the team approves or signs, and every message, draft, approval and signature is logged as evidence for the register.6
Summary
Keep one register of every AI tool, each with a written intended use, a risk rating, an owner and a person who checks the output, plus an incident route that can switch any tool off. Singapore's AIHGle 2.0 sets this out for deployers, and it maps onto the WHO principles and the MHRA's rules for software as a medical device.1
Each new AI tool goes live in an afternoon with its owner, its human check and its off switch already agreed. Singapore's 2026 guidance puts one register at the centre of AI governance, and the same page covers the scribe, the front desk drafts and the evidence answers in the consulting room.
See it in your clinicThirty minutes on your own patients and doctors.
Request a demoSee a note, a letter and a front desk reply drafted, approved and logged, ready for your clinic's AI register.
| Tool and intended use | Category (AIHGle 2.0) | Human check before use | Permitted data | Owner |
|---|---|---|---|---|
| Ambient note, drafts the consultation record | Clinical-Ops | Treating clinician reviews and signs | Consultation audio and the patient's record | Clinical safety lead |
| Referral and patient letters, drafted from the signed note | Clinical-Ops | Clinician reviews and signs | Signed note and letter templates | Clinical safety lead |
| Cited evidence answers during the consultation | Clinical | Clinician weighs the answer and decides | The question, with minimum patient detail | Clinical safety lead |
| Coding suggestions from the note | Ops | Clinician or billing lead confirms each code | Signed note | Front desk lead |
| Front desk reply drafts on every channel | Ops | Team member approves before it sends | Name, booking, prices and thread history | Front desk lead |
| Missed-call text | Ops | Team member approves the drafted text | Caller number and clinic hours | Front desk lead |
Governance covers every tool
AIHGle 2.0 separates Clinical use, where AI supports clinical decisions, from Clinical-Ops use, where AI sits in a clinical workflow without making the clinical judgement, and from Ops use such as scheduling.1 Its worked Clinical-Ops example is a tool that transcribes a consultation and summarises it into case notes for the doctor's review.1
Human oversight is mandatory for all Clinical and Clinical-Ops use, and deployers should not deploy AI that makes clinical decisions without a human making the final decision.1 That matches the first of the WHO's six principles, that humans remain in control of health-care systems and medical decisions.2 Across 67 countries reporting to WHO in 2023, 56% had no functional digital health governance.4
The AI tool register
AIHGle 2.0 asks organisations to keep and monitor a registry of all AI solutions they deploy, with each one's risk classification, so monitoring and mitigation are calibrated across the tool's life.1 For a private clinic that is one table with a row per tool, set out below.
It also asks organisations to specify the permissible data types for each solution.1 A booking drafter needs names, times and prices, and its row records that it receives no full clinical history.
Intended use, written before go-live
The MHRA publishes guidance for manufacturers on writing an intended use statement for software as a medical device, because intended purpose decides how software is regulated.3 Keep the manufacturer's statement and add one sentence on how the clinic uses the tool and who checks the output.
"Clinicians use the ambient note to draft the consultation record, and the treating clinician reviews and signs every note." Attach the supplier's written regulatory position to the same row.
A named safety lead and an owner per tool
AIHGle 2.0 notes that in a solo practice approval of an AI tool may sit with the business owner or clinical lead, and in larger organisations with the board or medical director.1 Deloitte's 2026 outlook advises health systems to assign ownership for each AI use case.5
The clinical safety lead is a registered clinician who signs off each clinical row and leads incident review. Each tool also has an owner who keeps its row current. The UK clinical safety officer role is covered in our guide to clinical safety standards for AI scribes.
Under the WHO principle of responsibility and accountability, stakeholders make sure AI is used under appropriate conditions by appropriately trained people.2 Record on the register who has been trained on each tool, and when.
An incident route that can switch a tool off
AIHGle 2.0 asks organisations to train staff to detect adverse events and respond in time, and to keep contingency plans that include shutting the AI solution down and switching to a protocol without it.1 Give every row a fallback, such as typing the note by hand or answering the inbox manually, and a named person who makes that call.
In Singapore, adverse events from medical devices go to HSA and data incidents to the PDPC.1 In the UK, the MHRA asks for problems with software as a medical device to be reported through the Yellow Card scheme.3 List the external route beside each tool.
Inside the clinic, a five-line form records the tool, the record, what happened and whether anything reached a patient, and the safety lead reviews the forms monthly.
How wai fits the register
wai is one clinical system, so the clinic registers one supplier across several rows. wai drafts the note, the letter, the codes and the front desk replies, and a clinician signs or the team approves before anything is saved or sent.6
Every message, draft, approval and signature is written to an event log the clinic can export, which feeds the monthly review.6 Role-based access is enforced in the database, so each row's data column matches who sees what.6 wai hosts each clinic's records in its own country, under that country's rules. We set this up with your clinic.
Where to start
List every AI tool in use today, including any staff adopted on their own. Give each a row with its intended use, category, human check, permitted data and owner, and name the clinical safety lead.
Write the fallback for each row, book a thirty minute monthly review for the first quarter, and file the register with the procurement checklist.
Bring your current list of AI tools and see wai's drafts, approvals and event log mapped onto your register.
Frequently asked questions
Does a small private clinic need an AI register?
Yes, and one page is enough. AIHGle 2.0 asks every deploying organisation to keep a registry with each tool's risk classification, and notes that in a solo practice approval can sit with the owner or clinical lead.1
Is a front desk AI tool part of clinical governance?
Yes, with a lighter rating. AIHGle 2.0 classes scheduling support as Ops use, and the risk depends on the human check and the data the tool receives.1 A person approves every reply.
Who should be the clinical safety lead?
A registered clinician with time set aside to review hazards, sign off clinical rows and lead incident review. In a small clinic this is often the medical director.
What counts as an AI incident?
Any output that was wrong, went to the wrong place or nearly reached a patient unchecked. Log near misses too, such as a wrong drug name a clinician caught in a draft, because they show where training is needed.
Is an AI scribe a medical device?
It depends on its intended purpose and the jurisdiction. In the UK, start with the MHRA's guidance on software and AI as a medical device, and get the supplier's position in writing.3 Our clinical safety guide covers the UK standards.
How often should the register be reviewed?
Monthly for the first quarter after any new tool, then quarterly, and whenever a tool or workflow changes. AIHGle 2.0 asks deployers to review performance periodically across the tool's life.1
Sources
- 1Ministry of Health Singapore and Health Sciences Authority, 2026Artificial Intelligence in Healthcare Guidelines (AIHGle 2.0), version 2.0Read 14 October 2026Read source
- 2World Health Organization, 2021WHO issues first global report on Artificial Intelligence in health and six guiding principles for its design and useRead 14 October 2026Read source
- 3Medicines and Healthcare products Regulatory Agency, 2026Software and artificial intelligence (AI) as a medical deviceRead 14 October 2026Read source
- 4World Health Organization, 2024Global strategy on digital health 2020-2025, report by the Director-General (EB156/35)Read 14 October 2026Read source
- 5Deloitte, 20252026 Global Health Care OutlookRead 14 October 2026Read source
- 6wai, 2026Security at waiRead 14 October 2026Read source
Adopt each new AI tool with confidence, owned and on the register.
Fifteen minutes with a clinician who uses wai daily, on how drafts, approvals and the event log fit your governance. We reply within one working day.
Partner with us
See it on your own patients, then build it with us
Thirty minutes on your own patients, doctors and rules, in your market. We reply within one working day, and the clinics that partner with us early shape what we build next.
See what wai does between the visits