What Tavrik means for you
Your hospital has put a safety net between you and the AI tools you use at work. It is called Tavrik. This page explains what it does, in about ninety seconds.
What changes for you
Almost nothing. You use ChatGPT, Copilot or Gemini the way you always have.
One thing is different: if you paste something with a patient’s details in it, those details are swapped for a label before the AI tool sees them.
What that looks like
Say you paste a handover note:
Mrs Eleanor Whitfield, MRN 4471902, DOB 12/03/1948, presented with…
The AI tool receives:
{NAME_1}, MRN {MRN_1}, DOB {DOB_1}, presented with…
The AI answers normally. When the answer comes back to you, the real names and numbers are put back in. You see your patient’s details. The AI company never did.
The labels look like {NAME_1} or {healthcare/medicare_number_au_1}. The word
in the braces tells you what was swapped. The number keeps them straight, so
{NAME_1} is the same person everywhere in your text.
Why your hospital does this
Patient details sent to an AI company leave the hospital’s control. Once they have gone, they cannot be recalled. This keeps them here, without stopping you using the tools.
What is recorded
Not what you typed. Your prompts and the AI’s answers are not stored.
What is recorded is a short line saying that a request happened: which department, which AI tool, what kinds of details were found, and whether it was allowed. That is what lets your hospital prove to an auditor that patient data is being protected — without anyone reading your work.
Nobody reviews your prompts. There is no transcript to read.
If something looks wrong
- The AI’s answer has a label in it, like
MRN_1without the braces. Rare, and harmless — nothing leaked. Ask your IT service desk about it. - You are asked to enter a pairing code. Your access has expired, usually after a device has been idle. Your IT service desk issues a new code; it takes a minute.
- Something was blocked. Your hospital sets rules about which tools may be used and what may be sent. If a request is refused, that is the rule, not a fault.
Who to ask
Your IT service desk or the person who set up AI tools for your department. They can see what happened to a specific request and explain it.
They cannot see what you typed. Nobody can. That is the point.