Intelligence should strengthen human judgement.

AI can help organise complex evidence. It does not take responsibility for a clinical decision. Qualified professionals remain responsible for care, and research teams remain responsible for how a system is designed, evaluated, and used.

SIPL’s public website and LifeBack preview do not provide diagnosis or treatment. Any future clinical workflow needs a clearly defined role for the system and the people who oversee it.

A model should be evaluated for its specific intended use before it informs a real workflow. Evaluation should test data quality, performance, limitations, and behaviour across the populations and settings where it may be used.

We do not treat a promising result in one dataset as proof that a system is ready for every clinic or population.

Users need to understand what information contributes to an output, what the output can and cannot say, and when evidence is incomplete. Poor-quality or missing inputs should not be presented as certainty.

Research involving health, behavioural, or genomic information requires appropriate permission, governance, and safeguards. Data should be handled for a clear purpose, with access limited to the people and services that need it.

Population representation is part of model quality. We aim to identify where evidence is limited and where further validation is needed rather than implying universal performance.

Responsible use does not end when a model is built. Research and deployments should have processes to review performance, investigate concerns, correct problems, and pause use when the evidence no longer supports it.

These principles describe SIPL’s approach; they are not a claim of certification or a substitute for product-specific governance documentation.