What it will not do
It does not diagnose, interpret results or recommend treatment. It will not call a package right for someone because of their medical history, and it does not waive a consultation. That is a clinical and regulatory line.
The core of this line (no diagnosis, no reading of results, no treatment advice) is held outside the model: a change to the instructions that removes it is refused.
Why the line sits here: under Health Sciences Authority (HSA) rules, diagnosing, interpreting results or advising treatment is what would make software a medical device. The AI in Healthcare Guidelines (AIHGle 2.0), published by the Ministry of Health (MOH) and HSA in March 2026, also say AI that patients use directly should not produce output that needs clinical expertise to read safely, disclaimers or not.
Results, emergencies, consent
- Questions about a patient's own results go to your team. It never reads out another person's results.
- If a patient's words suggest urgency, it gives 995 or the nearest A&E first, before any link or price. It never judges how serious a symptom is.
- Booking for someone else is fine, but the person tested must know and agree. Mentions of abuse, assault or self-harm go straight to a person.
What it does not ask for
It never asks for an NRIC or FIN in chat. To find a booking it asks for the phone number on it and the date, and the date of birth only before moving or cancelling one.