Generative AI is now a fixture in allied health workflows. Speech pathologists, occupational therapists, physiotherapists, and support coordinators are using tools like ChatGPT, Claude, and Microsoft Copilot to draft assessment reports, prepare funding letters, summarise session notes, and translate clinical observations into plain-language documentation for participants and families.
This is not a passing trend. Used well, AI saves clinicians hours of administrative time each week and frees them to spend more of their day with the people they support. Used poorly, it creates regulatory, privacy, and professional risks that can put a participant's funding, a clinician's registration, and a provider's NDIS approval in jeopardy.
The question is no longer whether to use AI in clinical practice. It is how to use it safely, ethically, and in a way that holds up to scrutiny.
Here is what NDIS providers should be paying attention to.
The regulatory landscape is moving quickly
Five frameworks now bear directly on how AI is used in NDIS-funded clinical work. None of them existed in their current form three years ago.
Your professional body's position on AI
Speech Pathology Australia released its position statement Artificial Intelligence in Speech Pathology: Ethical Considerations in May 2024. The Australian Association of Social Workers, Occupational Therapy Australia, and the Australian Physiotherapy Association have published or signalled comparable guidance. The Australian Health Practitioner Regulation Agency (AHPRA) has issued cross-profession statements that apply to every registered allied health professional.
The common thread across all of them is this: the clinician retains full professional accountability for every claim, recommendation, and clinical statement in any document, regardless of whether it was drafted with AI assistance. AI output must be critically reviewed, not signed off as a formality. Automation bias, the tendency to accept plausible-sounding output without scrutiny, is now a named and recognised risk.
If you have not read your profession's current AI guidance, that is the first thing to do.
Privacy Act 1988 and the Australian Privacy Principles
Pasting a participant's name, NDIS number, diagnosis, or clinical history into a public AI tool is a disclosure of personal and sensitive health information. It engages APP 6 (use and disclosure) and APP 8 (cross-border disclosure), because most major AI platforms process data offshore, typically in the United States.
Two practical implications follow.
First, the participant's original consent to collection of clinical information almost certainly did not contemplate disclosure to a foreign AI provider. New, specific consent or a recognised exception is required.
Second, de-identification before submission is the standard mitigation. Doing it well enough to actually break re-identification risk, in a population as small and well-documented as NDIS participants, is harder than it looks. A unique combination of disability type, age, location, and funding category can re-identify a participant even after names and addresses are stripped.
NSW, Victoria, and the ACT have parallel state-based health records legislation that adds further obligations.
NDIS Code of Conduct
The NDIS Code of Conduct, administered by the NDIS Quality and Safeguards Commission, applies to every registered and unregistered provider and worker delivering NDIS-funded supports. Its obligations of honesty, integrity, transparency, and acting with care apply directly to clinical documentation.
The Commission issued an AI transparency statement in February 2026, signalling growing scrutiny. A letter or report presented as the clinician's professional opinion but substantively AI-generated, without adequate review, could be characterised as misleading conduct. That is true even if the document is factually correct.
NDIA audit and fraud detection
The National Disability Insurance Agency has become noticeably more alert to AI-generated allied health reports as part of its fraud detection and assurance work. Documents that read as templated AI output, with generic clinical reasoning and recycled phrasing, carry reputational and audit risk even when the underlying clinical view is sound. Providers who lodge volume-driven, AI-pattern documents are increasingly likely to find themselves in a compliance review.
This matters in particular for letters supporting access to assistive technology, AAC devices, and other higher-cost supports. If you are unfamiliar with the formal pathway for these requests, our guide to low-cost AT and Replacement Supports sets out how the process works in practice.
Professional indemnity insurance
Most professional indemnity policies were written before widespread generative AI use. Some now contain explicit AI-use clauses, conditions, or exclusions. Before relying on AI for any document that supports a funding decision, confirm with your insurer whether your cover responds to a complaint or NDIA dispute concerning an AI-assisted document, and whether disclosure of AI use is a condition of cover.
Five principles for safe and defensible use of AI
The professionals we see using AI well in NDIS practice tend to operate on the same set of principles. None of these are technical. All of them are about how the tool fits into a clinical workflow that can withstand scrutiny.
1. You own every claim
If your name and provider number sit at the bottom of the document, you are professionally accountable for every word. AI output is a draft, not a finding. Read every clinical statement, check every measurement and date against your source records, and reword anything that does not reflect how you would actually express the view in your own clinical voice.
2. Strip identifying information before disclosure
Before any participant-related content goes into a public AI tool, remove names, addresses, dates of birth, NDIS numbers, school names, employer names, treating practitioner names, and identifiable family details. For small populations or rare conditions, consider whether the combination of remaining details could still identify the participant.
Better still, use enterprise-grade AI tools that offer Australian data residency, no-training guarantees, and auditable logs. The free public versions of these tools are not built for clinical data.
3. Get participant consent and document it
Consider updating your standard service agreement and consent forms to specifically reference AI-assisted documentation. Explain in plain language what AI is used for, what data is shared, what protections are in place, and what choices the participant has if they prefer their information not be processed by AI. Record consent in the file.
Participants and families are increasingly asking the question themselves. Having a clear answer is better than improvising one mid-session.
4. Disclose AI assistance in the document where appropriate
For letters and reports submitted to the NDIA, particularly Replacement Support applications, Section 10 substitution requests, and assistive technology assessments, consider including a short disclosure statement noting that AI was used to assist with drafting and that the clinician has reviewed and confirmed every clinical statement. This is consistent with the direction the Commission is signalling and is straightforwardly defensible if the document is later audited.
5. Build an audit trail
Keep your source clinical notes, the AI-generated draft, and the final reviewed version. If a document is ever questioned, the ability to demonstrate that the clinical reasoning originated with you and the AI was used as a drafting aid, not a reasoning engine, is what separates a defensible workflow from a problem.
Where AI genuinely helps
Within these guardrails, AI does meaningful work for allied health professionals. The use cases that consistently deliver value include:
Plain-language translation. Converting clinical findings into language a participant or family member can understand without losing accuracy.
Document structure. Reorganising existing content into NDIA-preferred structures for funding submissions, AAT preparation, and Replacement Support letters.
Note tidying. Cleaning up dictated or rough session notes into structured progress notes, working only from the clinician's own observations.
Letter drafting from existing reports. Producing a first-pass funding letter from clinical reports the clinician has already authored, with the clinician then reviewing every assertion against source. This is particularly common in AAC assessments and in OT functional capacity reporting, where the underlying clinical work is already substantial.
Research and orientation. Summarising current evidence on a particular intervention, condition, or product to support clinical reasoning, with the clinician verifying any cited source before relying on it.
Where to be careful
The use cases where AI introduces the most risk are the ones where it is asked to generate clinical reasoning rather than restate it. Asking an AI to justify a recommendation, infer functional impact from a thin set of observations, or invent the evidence base for a particular intervention crosses the line from drafting assistance to clinical fabrication.
Two specific patterns to avoid:
Do not ask the AI to write the rationale for a funding decision and then sign your name to it. The Replacement Support pathway, in particular, rests on the premise that an appropriately qualified treating professional has formed a clinical view that the replacement addresses disability-related needs and represents value for money. If the AI generates the reasoning rather than the clinician forming and articulating it, the letter may be technically accurate but substantively misrepresent how the conclusion was reached.
Do not let AI invent citations, statistics, or measurement values. Hallucinated detail is the most direct route to a fraud finding, and it is also the easiest thing for an auditor to spot.
The bottom line
AI in clinical practice is not the regulatory minefield some of the early commentary suggested. It is also not a free productivity dividend. It is a tool that, used with the same professional discipline you bring to any other clinical resource, can give you back hours of administrative time and improve the consistency of your documentation.
The clinicians who get this right treat AI the way they would treat a graduate clinician drafting on their behalf: useful, capable, in need of supervision, and never the source of the clinical view. The ones who run into trouble treat AI as a finished-product machine and discover the limits of that approach the first time a document is audited.
Frequently asked questions
Can NDIS providers use AI for clinical documentation?
Yes, with safeguards. AHPRA, Speech Pathology Australia, and other allied health bodies have all confirmed AI can be used to support clinical documentation, provided the clinician retains professional accountability, reviews every clinical statement, protects participant privacy, and meets their NDIS Code of Conduct obligations.
Do I need participant consent to use AI for their documentation?
In most circumstances, yes. Disclosing identifying clinical information to an offshore AI provider engages APP 6 and APP 8 of the Privacy Act 1988 and is unlikely to be covered by a participant's original consent to clinical record-keeping. Updating service agreements and consent forms to specifically address AI-assisted documentation is the standard mitigation, alongside de-identification where possible.
Will the NDIA detect AI-generated reports?
Increasingly, yes. The NDIA has become more alert to AI-generated allied health documentation as part of its assurance and fraud-detection work. Templated phrasing, generic clinical reasoning, and hallucinated citations are pattern signals that auditors look for. A document that originates with the clinician's own reasoning and uses AI only for drafting will read very differently to one generated end-to-end by an LLM.
Should I disclose AI use in my reports?
For documents submitted to the NDIA, particularly those supporting funding decisions, a short disclosure statement is increasingly considered best practice. It is consistent with the NDIS Quality and Safeguards Commission's February 2026 transparency signal and is straightforwardly defensible under audit.
Talk to our team
Assistive Tech Australia is a registered NDIS provider working with allied health professionals across the country every day. If you are an NDIS provider thinking about how to bring AI into your clinical or operational workflow, or working through an assistive technology recommendation that needs documentation support, get in touch with our team. We are glad to share what we are seeing across the sector.
This article is general information for NDIS providers and allied health professionals. It is not legal, professional, or compliance advice. Clinicians should consult their professional body, registration authority, employer, and professional indemnity insurer before adopting AI in their practice.
