AI Policy for Aged Care: What to Cover and What Staff Must Not Paste

An Australian aged care provider's AI policy should name approved tools, say who may use them for what, run a privacy impact assessment before any tool touches resident information, set out consent where capacity is limited, and keep an incident log. Staff must never paste identifiable resident, family or staff information into a public chatbot.

That last rule is the one that fails first. It fails because staff are busy, the chatbot is fast, and nobody told them where the line is. A policy that lists the tools and the never-paste items in plain words, on one page, will do more than twenty pages of principles.

Follow me on Instagram@sagnikteaches

Why the OAIC guidance is the starting point

The Office of the Australian Information Commissioner has published guidance on privacy and the use of commercially available AI products. Read the current page yourself; the points below are what matter for aged care, as of November 2026.

Connect on LinkedInSagnik Bhattacharya

• The OAIC says AI products should not be used simply because they are available. Decide first whether personal information is needed at all.
• Do due diligence on the product: whether it has been tested for your intended use, how people will oversee it, its privacy and security risks and who can access the data.
• Review the terms and settings for whether the developer can use your inputs for training or share data with third parties.
• Do a privacy impact assessment. The OAIC recommends privacy by design and a privacy impact assessment to understand and reduce the effects on individuals.
• Best practice is not to enter personal information, and especially sensitive information, into publicly available generative AI tools. Entering it can count as a disclosure, so the rules on use and disclosure apply, and sensitive information generally needs consent.
• Treat outputs as probabilistic, keep humans in the loop and record when information is AI-generated.
• Update privacy policies and collection notices, and tell people when AI may materially affect them.

Health information is sensitive information, and an aged care provider holds a great deal of it: diagnoses, medicines, behaviours, mobility, mental state, end-of-life wishes. That raises the stakes on every item above. This tutorial is practical guidance, not legal advice, so have your privacy officer or a lawyer read your final policy.

Subscribe on YouTube@sagnikteaches

The one-page structure that works

Staff will read a page. They will not read twelve. Build the policy in this order, and keep each section short enough that the page stays a page.

1. Purpose and who it covers (employees, agency staff, volunteers, contractors).
2. Approved tools and what each may be used for.
3. The never-paste list.
4. Consent and resident rights.
5. Human checking of outputs.
6. Reporting an incident.
7. Owner, review date and version.

The sections below give you wording and checks for each. Where a section says "verify", it names how you would confirm the policy is working in practice.

Approved tools: the short list that replaces guesswork

A policy that says "use AI responsibly" is not a policy. Replace it with a table of named tools. Illustrative example for a mid-sized residential provider:

Notice the third row. Saying "not approved" in the policy is better than saying nothing, because it removes the excuse that nobody said. Add a line telling staff how to request a new tool: a short form with the tool's name, the job it will do, the data involved and who benefits. Review each request against the same checklist, and write down the decision.

Verify the tool, not the brochure

For each approved tool, read the vendor's terms and privacy settings and record four facts: whether inputs are used for training, how long they are kept, where they are stored and whether a written agreement for confidential information is available. The privacy-terms checklist shows how. If you cannot find an answer, the tool stays on the "pending" line until you can.

The never-paste list: wording staff will remember

This is the heart of the policy. Write it as a list of things, not principles, and put it in a box.

NEVER paste or type into any AI tool that is not on the approved list:
- A resident's name, date of birth, room, photo or ID number
- Diagnoses, medicines, care plans, assessments or behaviour charts
- Incident reports, complaints or progress notes about a named person
- Family or representative names, contact details or conversations
- Staff names with performance, health, rostering or disciplinary details
- Anything from My Health Record or a clinical system
- Passwords, login codes or screenshots of internal systems
- Invoices or bank details for residents or families

Even on an approved tool: use the minimum, and never a full name.

Two points about the list. First, "remove the name" is not the test. A paragraph describing a resident with a rare condition, a distinctive family and a named room is identifiable without a name. The test is whether a colleague could recognise the person. Second, screenshots count. A staff member who photographs a screen of care notes and asks a chatbot to summarise it has disclosed the same information.

What staff can paste

Give them yes items too, so the policy is not just a list of bans. Public information, templates, generic scenarios, de-identified training cases made up for the purpose, a policy you are drafting, a roster pattern without names, an activity idea, a recipe for a texture-modified meal described in general terms. When people know what is safe, they stop guessing.

A privacy impact assessment in an afternoon

The OAIC recommends one. It sounds heavy, but for a single tool it can be a short document that a manager completes in two to three hours with the privacy officer. Cover these questions:

1. What job will the tool do, and could it be done without personal information?
2. What categories of information go in (resident, family, staff) and how sensitive are they?
3. Where does the data go, who can see it and how long is it kept?
4. Can the vendor use it for training? Can that be switched off?
5. What happens if the tool gives wrong output, and who checks?
6. Will residents or families be told, and how?
7. What is the plan if there is a breach, and how do we leave the tool and delete our data?

Illustrative: a provider wants to use a drafting tool for the monthly family newsletter. The assessment finds that the newsletter contains photos and first names from activities. The decision is that the tool drafts text only from a generic outline, photos and names are added by a person afterwards, and the tool never receives resident information. The PIA takes ninety minutes and the policy gains a row in the approved-tools table with the limits written in.

The same exercise on a tool that summarises care notes comes out differently. It touches health information, so the answer needs the vendor's written terms, a clear statement of where data is stored, consent arrangements and a trial on a small unit before wider use. Some tools will fail the assessment, and that is the point of doing one.

Consent where capacity is limited

The OAIC notes that sensitive information generally needs consent and that consent must be informed, current and specific. In aged care that is harder, because many residents have fluctuating or reduced capacity. Your policy should say, in plain steps:

• Staff do not assume consent from silence or from a notice on the wall.
• Where a resident has capacity, explain the use in simple words and record their answer.
• Where capacity is limited, involve the person's substitute decision-maker or representative as your state or territory rules and your own consent procedures require, and support the resident to take part as far as they can.
• Where a resident or representative declines, record it and make sure nothing about that person goes into an AI tool.
• Where AI is not used on resident information at all, say so in the policy. That is the simplest position and removes the consent problem.

The rules on who may consent for whom vary by state and territory, and by the resident's circumstances, so verify the arrangement with your legal adviser before you write specific wording. The policy should point to your existing consent procedure rather than invent a second one.

Human checking of anything the AI produces

The OAIC asks organisations to keep humans in the loop and treat outputs as probabilistic. For a care provider this means a named person reads and approves anything that affects a resident or goes outside the organisation. A simple rule set:

1. AI may draft; a person decides.
2. Nothing AI-written goes into a care plan, assessment or clinical note without a clinician reading every word.
3. Staff record when text was AI-drafted, in whatever note or file field you use.
4. AI never decides who is at risk, what medicine is given or whether an incident is reportable.

Before and after, illustrative: a manager asks an approved tool to draft a polite reply to a family complaint about laundry. The first draft apologises for "the incident of lost clothing on Tuesday" and promises a reimbursement the facility has no policy for. The manager deletes the promise, corrects the date and sends a shorter reply. The tool saved ten minutes. The check prevented a commitment nobody had approved.

An incident log that people actually use

Things will go wrong: a staff member pastes the wrong thing, a tool changes its terms, a draft sends an error to a family. A log helps you learn, and it matters legally. Keep it as a simple table with these columns: date, who reported, tool, what happened, what information was involved, who was affected, action taken, whether a notification was needed, date closed.

Under the Australian notifiable data breaches scheme, an organisation covered by the Privacy Act must notify affected individuals and the OAIC when a data breach involving personal information is likely to cause serious harm, as the OAIC's page describes. Your log should therefore have a column for the assessment decision and who made it, plus the date you first suspected a breach. The OAIC's data breach guide says that when you suspect an eligible breach you must take all reasonable steps to finish the assessment within 30 days, and it expects 30 days to be treated as a maximum, not a target; once you conclude a breach is notifiable, the OAIC and the affected people must be told as soon as practicable. Aged care providers also run an incident management system and report reportable incidents to the Aged Care Quality and Safety Commission under the Serious Incident Response Scheme, all now under the new Aged Care Act, which started on 1 November 2025. Connect the AI log to that existing incident process instead of running a parallel one, and decide in advance who checks whether an AI-related event is also a reportable incident. See the AI incident response plan template for a general version.

Make reporting easy and blame-free. A staff member who admits pasting a care plan into a public chatbot is doing exactly what you want, and if they are punished the next one will keep quiet. Say so in the policy.

Staff, family and supplier information is covered too

Most of this tutorial is about residents, but your policy has to protect the other people whose information you hold. Staff files hold health conditions, leave reasons, performance notes and bank details. Families give you contact details and sometimes their own health circumstances while talking about a parent. Suppliers send contracts and pricing that are commercially sensitive. A manager who pastes a disciplinary file into a chatbot to tidy the wording has made the same mistake as a nurse pasting a care plan.

Add one line for each group to the never-paste list, and one rule for managers: HR matters, complaints and anything legal go through an approved tool or through a person, never through a free chatbot. A realistic case: a facility manager wants help drafting a warning letter. The safe route is to ask the approved tool for a generic letter structure with placeholders, then fill in the facts by hand in the HR system. That takes five minutes longer and keeps the file out of a public tool.

Roster and agency data deserve a mention as well. A roster with names, availability and shift patterns is personal information about each worker, and it is often the first thing a manager is tempted to paste when asked to fix a gap. Ask for the pattern in anonymised form (shift types, counts, required skills) and apply names in your own system once the plan is agreed.

Realistic ways staff get this wrong

Illustrative situations, each followed by what the policy should make easy.

Before them, a quick sum on the breach clock, because it surprises managers. Say a carer pastes a resident's medication list into a free chatbot on a Tuesday, mentions it to her team leader on Friday, and the team leader emails the privacy officer, who reads it after a long weekend. Nearly a week has passed before anyone able to assess the breach knows about it. The OAIC says whether an organisation is "aware" is a question of fact in each case, expects suspected breaches to be promptly identified and reported to the right people, and warns against unreasonable delay, so a week lost between junior staff and the privacy officer is hard to defend against a 30-day maximum. That is why the policy says "the same day".

1. The agency nurse. A nurse from an agency, unfamiliar with your rules, uses a personal chatbot app to rewrite a handover. Fix: the policy covers agency staff, and induction includes the one-page never-paste list.
2. The helpful translation. A carer pastes a family's message in another language into a free translator to reply. The message includes a resident's name. Fix: approved translation process, and a note that translators count as AI tools.
3. The photo. A staff member photographs a whiteboard in the nurses' station and asks an app to tidy the text. Fix: the never-paste list covers screenshots and photos.
4. The menu plan. A cook pastes dietary needs for residents by room number to draft a menu. Fix: use anonymous categories (texture-modified, diabetic) and numbers, not rooms.
5. The tool that changed its terms. A vendor adds a feature that shares data for improvement. Fix: a review date for every approved tool and someone assigned to watch vendor notices.

Rolling the policy out on the floor

Hand it out once, and it will be forgotten. Put the one-page never-paste list at the nurses' station and in the staff room, and cover it in the first ten minutes of each induction. At handover, a team leader can read one line aloud each week for a month. For a mid-sized provider the whole rollout takes about three hours of manager time to prepare and ten minutes per team.

Test it by asking three staff members, a fortnight later, what they must never paste into a chatbot and who to tell if they do. If they can answer both, the policy is working. If not, shorten it. The tutorial on rolling out an AI policy has more ways to make it stick, and the care home 90-day implementation plan shows where the policy sits in a bigger programme.

A filled-in extract of the policy page

Seeing finished wording helps more than a blank template. Illustrative extract for a mid-sized residential provider, written in the voice staff will read it in:

AI USE AT [PROVIDER NAME] - version 1.0, approved [date], next review [date]
Owner: Privacy Officer. Backup: Director of Care.

1. Who this covers: all employees, agency staff, volunteers and contractors.
2. You may use only the tools on the approved list, for the jobs listed beside them.
3. You must never put resident, family or staff information into a tool that is not on the list (see the box).
4. A person always reads and approves AI-drafted text before it is used. You are responsible for what you send.
5. If you think you have pasted something you should not, tell your manager the same day. You will not be in trouble for reporting.
6. Residents and representatives may say no to AI being used with their information. Record it and respect it.
7. Questions or new tool requests: [contact].

Seven lines is enough. Anything longer belongs in a supporting procedure that managers read, not on the page staff see. Notice that line 5 promises no blame. That sentence does more for your incident log than any penalty clause, because the cheapest way to learn about a leak is for the person who caused it to tell you.

How to check the policy is working

Policies fail quietly, so test them. Three simple checks, each taking under an hour a quarter:

1. The ask-three test. Ask three staff members from different shifts what they must never paste into a chatbot and who to tell. Record how many get both right.
2. The log review. Look at the incident log. An empty log after six months is not good news; it usually means nobody is reporting. A log with a few minor entries and quick actions is healthier.
3. The tool census. Ask managers which AI tools their teams actually use, including on personal phones. Compare the answer with the approved list. Every gap is either a tool to assess or a habit to stop.

If the ask-three test shows poor recall, shorten the policy. If the census shows heavy use of an unapproved tool, find out which job it is doing and whether an approved tool can do it. Staff rarely use a chatbot out of mischief; they use it because a task was slow.

Owner, review and what to record

Name one owner, usually the privacy officer or a senior manager, with a backup. Put the version number, the date approved and the next review date on the page. Keep a register of approved tools, the decision for each, the privacy impact assessment and the date each was last reviewed. Review every six months at first, and immediately after any incident or any change to a vendor's terms.

Finally, record the decision not to use AI where that is the decision. A provider that has assessed a tool and said no has done its job. If you would like to work through the approved list and never-paste list with someone, that is what my AI implementation consultation covers, alongside your own privacy officer and legal adviser, who must have the final word on the wording.

Questions aged care managers ask about an AI policy

Do we need an AI policy if we only use a free chatbot occasionally?

Yes, and a short one. Occasional use is when mistakes happen, because nobody has agreed what is allowed. A one-page policy listing approved tools, a never-paste list and who to tell if something goes wrong takes an afternoon to write and a few minutes to read.

Can staff paste a care plan into a chatbot if they remove the name?

Usually not. A care plan combines a diagnosis, a room, a family situation and behaviours that can identify a resident even without a name. Treat a care plan as identifiable. Use an approved tool or write a fictional version that cannot be traced.

Who should approve a new AI tool?

Name one person, such as the privacy officer or a senior manager, and one backup. The approval is a short written decision recording what the tool is for, what data goes in, the vendor's terms and the date reviewed. Staff should never approve their own tools.

How often should the policy be reviewed?

Every six months at first, then yearly, and straight away after an incident or when a vendor changes its terms. AI tools change quickly, so a policy last reviewed two years ago is probably out of date.

Further reads

• AI Consultant for a Care Home: Do You Need One, and What to Ask — Whether you need outside help, and what to ask.
• Can AI Write a Care Home Incident Report? Template and Checks — Incident wording and checks, with AI kept in its place.
• AI for Disability Services: Service Agreements and Incident Logs — Agreements and incident logs in a related care setting.
• Care Home Daily Notes: Better Wording With AI, Checked by Staff — Daily note wording with staff review.
• Care Home Medication Audit: A Monthly Template and AI Action Log — A monthly audit where staff may be tempted to paste data.
• Care Home Mandatory Training Tracker: Renewal Dates and AI Overdue Alerts — Where policy training can be tracked for each staff member.
• NDIS Provider Policies and Procedures: Tailoring Templates With AI — A policy list, tailoring prompts, a review log and a staff training record for turning a template into your own NDIS policy, with an adviser's check.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Office of the Australian Information Commissioner guidance on privacy and commercially available AI products, its notifiable data breaches pages and its data breach preparation and response guide (fetched 4 November 2026); Department of Health page on the new Aged Care Act and the Aged Care Quality and Safety Commission's Serious Incident Response Scheme page (archived copies from June and July 2026, read 4 November 2026).