Admission of a new tool
Every new AI tool comes in through one front door.
No forms passed around by email and no tool switched on before it is reviewed. Here is what a hospital sees when someone asks to bring a tool in.
01
One request
A clinical or AI leader names the tool, its vendor, what it is for and the clinical workflow it will serve, and answers five plain questions about what it will touch.
02
The vendor's documents
Attach up to 50 PDF, Word (.docx), PowerPoint (.pptx), Markdown or text files, 25 MB each. Each is checked for patient names and numbers on record before private storage. Legacy Word files must first be saved as .docx.
03
A first risk tier
Routine, Sensitive or Restricted, set by fixed published rules from the five answers, with the reasons shown. No model reads the documents to decide.
04
A named decision
The Chief AI Officer, a Model Approver or the Clinical Governance Lead confirms or changes the tier and accepts or returns the request, with a note. Nobody decides their own request.
05
Into the tool record
An accepted tool is registered and goes through the hospital's model approval before any clinician can use it. Every step is a row on the hospital's ledger.
The five questions
- Does it act without a clinician signing off?
- Does it read patient data?
- Does it send anything to a cloud service?
- Does it inform clinical decisions?
- Does it produce content patients see?
How the first tier is set
- Restricted
- Acts without sign-off, sends patient data to a cloud service, or produces clinical content patients see.
- Sensitive
- Reads patient data, runs in a cloud service, informs clinical decisions, or faces patients.
- Routine
- None of the above.
The tier is a starting point; the deciding leader confirms or changes it, and the reason is kept.
The AI assists; clinicians decide.