Not all AI use has the same risk.
Using AI to draft a generic handout is different from using AI to transcribe a real consultation. Summarising a public article is different from uploading a referral letter. Creating a fictional teaching case is different from asking a tool to help write a clinical recommendation.
That is why allied health teams need a simple way to classify AI workflows before they choose tools or enter information.
One practical model is green, yellow, red.
It is not a legal test. It is a team conversation tool. It helps people slow down and ask: what information is involved, what is the task, what tool are we using, and what could happen if the output is wrong?
Green: lower-risk starting points
Green workflows use public, fictional, general, or non-sensitive information. They usually involve drafting, structuring, brainstorming, simplifying, or preparing first-pass material.
Examples:
- drafting a generic patient education handout
- creating a fictional case for teaching
- planning a staff in-service
- rewriting public information in plain English
- creating a meeting agenda
- drafting non-sensitive admin text
- making a checklist for a common appointment type
- summarising public research abstracts for later review
Green does not mean “no review needed”. It means the workflow is a sensible starting point because it does not require real client information and does not ask AI to make a clinical decision.
The usual controls are simple:
- keep real client information out
- use public or fictional input
- review the output before use
- check sources if the output includes clinical or research claims
- make sure the final material fits your audience and setting
Green workflows are a good place for teams to build AI literacy.
Yellow: needs more thought before use
Yellow workflows may involve personal information, sensitive information, uploaded files, client-facing output, workflow impact, or a higher consequence if the output is wrong.
Examples:
- AI-assisted report drafting
- client-facing letters or emails
- uploaded referral documents
- AI scribes or transcription tools
- summaries of real consultations
- rewriting clinical notes
- marketing claims about services or outcomes
- policy documents that affect staff or clients
- tools connected to practice systems
Yellow does not mean “never”. It means the team should pause and review the workflow properly.
For yellow workflows, ask:
- What information goes into the tool?
- Is any of it personal, health, staff, or confidential information?
- Is the tool approved for that information?
- Where is the information stored or processed?
- Is consent needed?
- Who reviews the output?
- Could the output affect care, records, funding, referrals, rights, or client expectations?
- What happens if the output is wrong?
Yellow workflows often need local approval, clearer staff instructions, and a documented review process.
Red: do not use this way
Red workflows are uses that should not happen in public, personal, or unapproved AI tools.
Examples:
- identifiable client information in a public chatbot
- clinical notes, reports, photos, recordings, or transcripts in an unapproved tool
- complaints, incidents, staff matters, contracts, or confidential business information in a personal AI account
- asking AI to make clinical decisions
- asking AI to decide risk, diagnosis, treatment, access, funding, or eligibility
- using AI output directly in a clinical record without review
- recording or transcribing consultations without the required explanation, consent, and local approval
Red means the workflow is outside the agreed boundary. It may need a different tool, a formal review, a clinical governance process, legal or privacy advice, client consent, or a decision not to use AI for that task.
For staff, the rule should be clear: if the workflow is red, stop and ask before proceeding.
The colour can change when the details change
The same tool can be green, yellow, or red depending on the workflow.
A public chatbot used to draft a general handout may be green.
The same chatbot used with a real referral letter may be red.
An AI scribe may be yellow if it has been reviewed, approved, explained to clients, and used with a clear consent and review process.
The same scribe workflow may be red if staff start recording consultations through an unapproved account without telling clients what is happening.
This is why “Can we use AI?” is too broad. Teams need to name the tool, the task, the information involved, and the controls around it.
Use the model before choosing a tool
Many teams choose the tool first. The safer sequence is:
- Name the workflow.
- Classify the information.
- Classify the workflow as green, yellow, or red.
- Choose a tool that fits that risk level.
- Decide who reviews the output.
- Write down the local rule.
- Review the workflow after real use.
This prevents a common mistake: buying or adopting a tool because it looks impressive, then trying to work out the safety process later.
A simple team exercise
In a team meeting, write down ten ways staff might use AI.
For each one, ask:
- Is the input public, fictional, general, personal, health-related, staff-related, or confidential?
- Is the output internal, client-facing, clinical, educational, administrative, or operational?
- Could the output affect care, records, funding, communication, rights, or trust?
- Is the tool approved for the information?
- Who checks the output?
Then label each workflow green, yellow, or red.
This exercise usually makes the real issue clearer. Instead of talking about AI as one big category, the team can separate the easy starting points from the workflows that need review, and from the uses that should stay out of the tool being considered.
Where the resources fit
If your team is just starting, use the Before You Paste Into AI checklist before entering information into a tool.
If you want a small set of resources to help with the first team conversation, use the AHAI AI Privacy Starter Pack.
If you want editable templates, policy material, tool assessment checks, consent wording examples, training slides, and implementation resources, see The Allied Health AI Safety Kit.
Source note
This article is a practical educational resource, not legal or clinical advice. It is informed by Ahpra guidance on AI in healthcare, OAIC guidance on privacy and commercially available AI products, and Australian Government information on AI in health care.