What to take out of an NDIS case note before you use AI

A shift note gets written at the end of a long day, often on a phone, and a chatbot will happily tidy it up. The trouble is what the note carries: a participant’s name, their NDIS number, their medication, their housemate’s bad night.

This page covers what comes out before any of that goes near an AI tool, where NDIS rules say something and where they’re silent. Which accounts hold up under the Privacy Act is a separate question, covered in choosing an AI tool under the Privacy Act.

Can support workers use ChatGPT for progress notes?

No NDIS rule names a chatbot and bans it. The NDIS Code of Conduct applies instead, and to you personally: section 5 of the Code of Conduct Rules covers providers and anyone they employ or otherwise engage. Section 6(1)(b) says you must respect the privacy of people with disability.

The Commission’s own AI document, a position statement from February 2026, is about behaviour support plans. It doesn’t endorse AI tools for them and expects participant information to be de-identified before any AI use, a word the Privacy Act defines and de-identified versus pseudonymised unpacks.

The worker carries the duty, but the provider picks the tools. A personal login on your own phone is the version I trust least, because nobody at work knows what it keeps. Walk your own setup through this:

What to take out of an NDIS case note: one shift note, checked

Below is an invented shift note, the kind written at 9 pm in a supported independent living house. The box runs our Australian text checker in your browser. It flags the NDIS number by format alone, nine digits starting with 43, with no check digit to confirm it.

It also catches the mobile and the email. Now count what it leaves: the name, birth date, street, epilepsy medication, housemate, Mum, GP and day program. None has a fixed shape, and together they’re most of the note. The wider list of personal data to redact sorts them by type.

After a full pass by hand, Kirra becomes “the participant, a woman in her thirties”, the street goes, the epilepsy medication becomes “her regular evening medication” and Dr Pillai becomes “the GP”. What stays is what an AI needs: what happened, in what order, and what the worker did.

The pass took out more than numbers, and that’s the part people skip. If notes leave your system as a spreadsheet, masking a CSV export column by column is the bulk version of the same job.

Third parties and places that identify a participant

A note about one participant is usually a note about five people. The housemate who shouted, the mum who rang, the GP, the worker on the next shift. Each has their own privacy, and a pasted note hands a chatbot their name beside a bad evening.

Places finish the job. “The Gidgee bus ran late” in a town of 900 people is a name with extra steps. The Privacy Act asks whether someone is reasonably identifiable, which de-identified versus anonymised sets beside the overseas terms.

No checker does this step, and I wouldn’t let one try. Read the cleaned note as the neighbour who works at the chemist would read it: if she’d know who it is, it isn’t clean yet.

Are disability support notes health information under the Privacy Act?

Section 6FB counts assessing, maintaining, managing or recording someone’s health as a health service, and subsection (3)(b) says that still holds in the course of caring for a person with a disability. Personal information collected in providing that service is health information under s 6FA, a kind of sensitive information.

Not every shift is a health service, but I treat any note recording medication, seizures, behaviour or mood as one, and then even its address line counts. Turnover won’t help: s 6D(4)(b) removes the small business exemption for anyone who provides a health service and holds health information.

Victoria adds a layer: the Health Records Act 2001 and its 11 Health Privacy Principles. HPP 9 is written for health information going outside Victoria, so a Sydney data centre can raise it, not only an overseas one. Registered therapists also answer to Ahpra’s AI guidance.

Is it ethical to use AI for case notes?

Privacy is half of it. The other half is whether the note is still true. Clause 12 of the core Practice Standards wants each participant’s information identifiable, accurately recorded, current and confidential, and an AI rewrite can quietly change what happened.

A tidier note that says Kirra “refused” when she said “not now” is a worse record, and it’s the one that gets read at a plan review. Smoothing exactly those words is what a chatbot does best.

My line: AI can fix grammar and structure on a cleaned copy, and the facts, the participant’s own words and the signature stay with the worker. If a full note has already gone in, whether that’s a notifiable breach is the next question.

Automated decisions and NDIS privacy policies from 10 December 2026

From 10 December 2026, APP 1.7 to 1.9 apply to organisations covered by the Privacy Act. If you’ve arranged for a computer program to make a decision, or do something substantially and directly related to one, using someone’s personal information, your privacy policy may need to say so.

The trigger is a decision that could reasonably be expected to significantly affect someone’s rights or interests. One of the Act’s examples reads like the NDIS: access to a significant service or support. A shift note tidied by a chatbot isn’t a decision; a referral triage score could be.

The policy then lists the kinds of personal information used and the kinds of decisions involved. The OAIC published its guidance on 30 September 2026. A borderline tool is better listed than argued about later, and our AI policy template is a start for the rest.

An AI policy for NDIS providers: the checklist

Support workers do what’s easy at 9 pm. The provider’s job is to make the safe version the easy one, then write it down so it holds up at audit and in a complaint. Staff files need the same care: a TFN in a payroll export is its own problem.

Two rules carry most of the weight: which accounts are allowed, and who reads the cleaned note for people and places before it goes. Approving one tool properly beats pretending nobody uses any.

Cleaning an NDIS case note before AI, on your own computer

Numbers and contact lines are the tedious part of cleaning a case note, and the part software should do. Nonimo for Australian teams takes names, NDIS and Medicare numbers and contact details out of a note on the worker’s own computer before it reaches an AI tool.

When the reply comes back, the original names and numbers return to their places on that screen. The housemate, the town and the bus route are still a person’s call. I’m reading public rules here, not your service agreement or your registration conditions.

Sources

• NDIS Quality and Safeguards Commission, Position statement: Use of artificial intelligence in the development of behaviour support plans. February 2026: behaviour support plans, no endorsement of AI tools, de-identification before use.
• National Disability Insurance Scheme (Code of Conduct) Rules 2018. Section 5 on who is covered, s 6(1)(b) on privacy, and the s 73V civil penalty note.
• Privacy Act 1988. Sections 6FA, 6FB(3)(b) and 6D(4)(b); APP 6.1.
• OAIC, Guidance on privacy and the use of commercially available AI products. Keep personal and sensitive information out of public generative AI; describe AI use in privacy policies.
• OAIC, New resources on transparency for use of AI and automated decision-making. APP 1.7 to 1.9 from 10 December 2026; guidance published 30 September 2026.
• Health Complaints Commissioner (Vic), Health records: providers. The 11 Health Privacy Principles, including HPP 2 and HPP 9 on information leaving Victoria.
• Ahpra, Meeting your professional obligations when using AI in healthcare. Informed consent before patient data goes into an AI tool.

Accessed 7 November 2026.