Ask
Say who you are and what is in front of you. You will get questions to take to a visit, a huddle, a pitch, or a kitchen table.
A health system is not a market first. It is people who will live with what you ship. Translation is fitting the tool to that place.
For I am building or bringing a tool into care
Opening
What would we need to know before we let this rank, write, or watch a person in care?
Who was this tested on, and do they look like the people who will use it?
What must remain a human judgement even if our model is faster?
If this fails, is it a nuisance, a delay, or an irreversible harm, and who is named?
What can a patient or a clinician refuse, without the rest of care getting worse?
Where does the data go, including audio and notes, and who is the controller?
Same questions, denser rooms. Go when you are ready — not as a test of seriousness.
Health-specific guidance
World Health Organization
WHO's 2024 guidance on large multimodal models in health: consent, evidence, equity, and the duties that remain with clinicians and states.
The question is not 'is AI risky?' but 'what does care require of this tool?'
How to manage, not only name
U.S. National Institute of Standards and Technology
A voluntary framework for putting AI risk into an organisation's ordinary risk practice. Widely referenced; not a law.
You are past 'what can go wrong' and need a cycle: govern, map, measure, manage.